a travel route recommendation system based on smart...
TRANSCRIPT
Research ArticleA Travel Route Recommendation System Based on SmartPhones and IoT Environment
Chenzhong Bin 12 Tianlong Gu3 Yanpeng Sun4 Liang Chang 2 and Lei Sun5
1School of Information and Communication Guilin University of Electronic Technology Guilin 541004 China2Guangxi Key Laboratory of Trusted Soware Guilin University of Electronic Technology Guilin 541004 China3Guangxi Experiment Center of Information Science Guilin University of Electronic Technology Guilin 541004 China4School of Mechanical and Electrical Engineering Guilin University of Electronic Technology Guilin 541004 China5School of Electronic Engineering and Automation Guilin University of Electronic Technology Guilin 541004 China
Correspondence should be addressed to Chenzhong Bin binchenzhong163com
Received 15 March 2019 Revised 26 May 2019 Accepted 1 July 2019 Published 14 July 2019
Academic Editor Gerardo Canfora
Copyright copy 2019 Chenzhong Bin et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Tourism recommendation systems play a vital role in providing useful travel information to tourists However existing systemsrarely aim at recommending tangible itineraries for tourists within a specific POI due to their lack of onsite travel behavioral dataand related route mining algorithms To this end a novel travel route recommendation system is proposed which collects touristonsite travel behavior data automatically regarding a specific POI based on smart phone and IoT technology Then the proposedsystem preprocesses the behavior data to transform raw behavior sequences into Tourist-Behavior pattern sequences Subsequentlythe system discovers frequent travel routes from the generated pattern sequences by using an original route mining algorithmnamed Tourist-Behavior PrefixSpan Finally a route-recommending method is designed to search and rank tangible travel routesaccording to the querying touristrsquos profile and constraintThe experimental results demonstrate that the proposed system is efficientand effective in recommending POI-oriented tangible travel routes considering touristsrsquo route constraints and personal profile whileensuring that the suggested routes have considerable route values
1 Introduction
Tourism is a popular leisure activity with the goal of visitingsome Points of Interests (abbr POIs) based on onersquos personalpreference and constraints Recently industry and academiahave been studying and developing tourism recommenda-tion systems for providing convenient travel information totourists including next POI suggestion [1ndash7] Top-k POIsrecommendation [8ndash10] and POIs travel route recommen-dation [11ndash18] Particularly travel route recommendations aremore practical and useful than the two former kinds of POIrecommendations in practice yet they are more challengingTravel route recommendation aims to organize a bundleof candidate POIs as a reasonable visit sequence (ie anitinerary) while adhering to personal constraints of a giventourist for example a limited time or finance budget user-specified start and end locations
Thus many researchers have studied travel route recom-mendation issues and designed various algorithms for solving
these problems [19]Most of these works are orienting to city-level or district-level itinerary recommendation scenario thatis planning a POIs travel route within a city or regionfor tourists On the other side it is difficult for touriststo instantly choose personal interested exhibits or spotsand arrange these items into an itinerary under their timebudget when travelling in a never visit POI for examplea museum or a park This situation makes most of touristsroaming or missing some potential interested items in thePOI However few existing systems manage to recommendtangible itineraries for tourists within a given POI due tolack of rich onsite travel behavior data and related itinerarymining algorithms
The popularity of smart phones and the flourish Internetof Things (IoT) techniques provide various means to senseonsite travel behaviors of tourists including not only travelspatial-temporal trajectories [20] and visit durations but alsothe tangible travel behaviors [21] such as taking photosstanding or walking It is a common sense that onersquos onsite
HindawiWireless Communications and Mobile ComputingVolume 2019 Article ID 7038259 16 pageshttpsdoiorg10115520197038259
2 Wireless Communications and Mobile Computing
travel behaviors imply his or her objective preferences andinterests to some objects For instance tourists will spendlonger visit duration or take more pictures or stand stillmore times to appreciate something on a spot if they aremore interested in somethingThus gathering touristsrsquo onsitetravel behaviors and mining their personal preferences andfrequent travel routes could be an effective approach forrecommending tangible travel routes for new similar touristsin specific POIs
To that end in this work we proposed a POI-orientedtravel route recommendation system based on IoT technol-ogy and smart phones In detail we first adopted Bluetoothlow energy (BLE) beacons [22] to periodically broadcastpositioning information for nearby smart phones Also wedeveloped a client App running on an Android smartphoneto collect onsite travel behaviors data and correspondingpersonal profiles and then upload collected data to thesystem server Next on the system server side all collectedtravel behavior data are classified according to their per-sonal profiles And then a behavior sequence preprocessingmethod and Tourist-Behavior pattern mining algorithmswere designed to generate diverse tangible travel routes Atroute recommendation stage to ensure the personalization ofour recommendations the proposed route ranking methodrecommends tangible travel routes for new tourists by usingtheir personal profiles and route constraints As all tangibletravel routes are constructed from real historical onsite travelbehavior the recommended routes have high accuracy andrationality in terms of visit arrangements Also since therecommended routes are retrieved from the correspondingcandidate route subset according to the querying touristrsquo sprofile the visit objects of the final route can suit the personalinterests of the tourist better
Our main contributions are summarized as follows (a)An onsite travel behavior data collecting method which isbased on touristsrsquo smartphones and Bluetooth low energy(BLE) beacons is designed to automatically sense onsitetravel behavior under indoor and outdoor tourism scenarios(b) Tourist-Behavior PrefixSpan algorithm is proposed togenerate diverse frequent travel routes effectively based onhistorical Tourist-Behavior pattern sequences (c) Travelroute ranking method is proposed to recommend a list oftangible travel routes according to the querying touristrsquosprofile and constraints so as to ensure the route value andrationality of the final travel routes (d) Experimental resultsdemonstrate the effectiveness of our system in recommend-ing personalized tangible travel routes for tourists in a givenPOI based on historical onsite travel behavior
The rest of the paper is organized as follows Section 2discusses the related works regarding travel route recom-mendation systems and tourism recommendationwithin IoTenvironment Section 3 presents the research methodologiesof the proposed system where the framework of the system isdescribed in Section 31 onsite travel behavior data collectingmethod is explained in Section 32 The Tourist-Behaviorpattern sequences mining method and tangible travel routerecommendation procedure are thoroughly explained in Sec-tions 33 and 34 respectively Section 4 analyzes the experi-mental results to validate the feasibility and performance of
the proposed system The conclusion and future works arepresented in Section 5
2 Related Works
21 Travel Route Recommendation Systems Due to the prac-tical values of travel route recommendation systems lots ofresearchers have been placing a great emphasis on solvingthe route planning in tourism scenarios [19 23] in recentyears One category of these works uses the Orienteeringproblem [24] and its variants to approach the route planningproblem These methods formulate the problem from differ-ent perspectives resulting in diverse problem models whichconsider different problem variables and constraints Theroute-generation process actually is a near-optimal solutionusing metaheuristic searching algorithm [25] Accordinglyto enhance the personalization of the recommended routesmost of these works resort to acquiring more detailed userfeedbacks or profiles to assist in fine-tuning the final resultsIn [26] the system solicits walking travel related attributesfrom tourists to insert concrete walking routes into POIitineraries thereby supporting more experiential explorationof tourist destinations Zhang et al [27 28] studied tourrecommendation with the goal of recommending person-alized itineraries based on the interest preferences of usersand available touring time while considering opening hoursof POIs and uncertainty in travelling time Other studiesconsider more practical factors that raise novel optimizationchallenges incorporating forms of situational awareness suchas multiple modes of transport [29] considering trafficconditions [30ndash32] POI crowdedness [33 34] and queuingtimes [35]
Although the optimization-based route planning systemscan recommend a reasonable travel route adhering to onersquospreferences and constraints the interactive preferences inputof the planning process is time costly to tourists It isimpractical for tourists to spend a long time in inputtinga complex user profile when entering a specific POI Fur-thermore the results of these systems are lack of diversityand less personalization due to the near-optimal solutionsearching methodology Therefore lots of other works focuson generating personalized travel routes by mining UserGenerated Contents (UGC) that is data-driven approachesto route planning The UGC adopted in previous researchesinclude GPS trajectory datasets [36] check-in datasets [3738] and geo-tagged photos [39]
Chen et al [31] adopted historical check-in data and GPStrajectories to construct a POI network and used a heuristicmethod to generate a favorite POIs list for a specific user in aninteractive manner Subsequently the system requests usersto specify their favorite POIs during the route-generatingstage PersTour system [13 40] uses geo-tagged photos todetermine POI and construct POI travel sequences andleverage an Orienteering problem solving model to recom-mend POI itineraries by both considering POI popularityand tourist personal interests Majid et al [41] inferred thelocation of POIs and their semantic meaning using clusteringapproaches on geo-tagged photos and used a pattern mining
Wireless Communications and Mobile Computing 3
algorithm to discover popular travel sequences under thecontext of the tour recommendation that is time day andweather Besides the rapid growth of online tourismwebsitesprovides massive POI reviews and travelogues Thus somerecent works [14 42ndash44] focused on generating personalizedmining travelogues and POI related contents
The data-driven based route planning systems can rec-ommend rather personalized and reasonable travel routeshowever the limitation of these systems is that they aimat recommending city-level or district-level orienting POIsitineraries that is planning POIs travel routes for touristswithin a city or region They fail to generate tangible travelroutes for tourists within a specific POI due to lack ofrich onsite travel behavior and related itinerary miningalgorithms
22 Personalized Recommendation in IoT Environment TheIoT concept was first coined by Kevin Ashton in 1999 [45]in supply chain management applications based on radiofrequency identification devices (abbr RFID) At present IoTis referring to a bundle of technologies that aim at sensinghandling and transmitting state information of physicalenvironments which is broadly applied in smart cities [4647] smart business [48 49] and smart tourism scenarios[50]The goal of these smart systems is to recommend a set ofpersonalized and valuable items or services for various usersTo this end researchers focus on recording and analyzinguser behavior to learn user preferences more precisely bydeploying IoT technologies
Specifically in smart tourism applications some studiesfocus on using IoT technologies and mobile devices toimprove tourism experiences in an interactive way Kuusiket al [51] designed a smart museum system that integratesPDAs and RFID technologies to provide users with culturalcontents by sensing the interactive behavior between PDAsand RFID tags which were installed near each artworkIn [52] an indoor location-aware system was designed fora smart museum to enhance visitorsrsquo cultural experiencesThe proposed system obtains visitors localization informa-tion through a Bluetooth low energy (BLE) infrastructureinstalled in the museum and uses several location-aware ser-vices hosted in the system to interact with visitors accordingto their locations
And some other works aim at solving the next visitingspot recommendation problem within a specific POI Mas-simo et al [5] leveraged Inverse Reinforcement Learningmethod to learning user preferences by observing touristsonsite behavior in an IoT-equipped smart museum so as topredict next exhibit sequentially for tourists Hashemi et al[6 7] solved the challenging next POI recommending prob-lem by logging and mining usersrsquo onsite physical and onlineinteraction behavior data within an IoT-augmentedmuseumHowever the above works fail to generate personalized andtangible travel routes for tourists
To that end some researchers strived to solve thischallenging problem by mining historical travel trajectoriesin an IoT-augmented environment Tsai et al [15] adoptedRFID infrastructure to record visitorsrsquo check-in sequencesof recreation facilities in a theme park and then proposed a
statistical method to find behavioral similar historical visitorsso as to suggest a travel route for the current queryingvisitor Luo et al [16] studied a new path finding system thatdiscovers the most frequent path during user-specified timeperiods in large-scale historical trajectory data Tsai et al[17] proposed a touring path suggesting system for visitorsto comprehend exhibits in exhibitions or museums Thesystem takes previous popular visiting trajectories as the sug-gestion foundation and provides a time-interval sequentialpatterns mining algorithm improved from [18] to generatepersonalized travel routes However as the above systemsonly resorted to dedicated IoT devices to record the check-in behavior they hardly learn more tangible user preferencestowards each interest object from the single dimensionalbehavior Meanwhile current smartphones generally equipa camera and diverse sensors which could be used tosense multiple dimensional onsite behaviors of tourists soas to explore high-level tourist preferences and recommendpersonalized tangible travel route Although some previousresearches have investigated the human activity recognitionbased on smartphone sensors [21 53] there is no study onlearning user preferences directly by smartphones To thebest of our knowledge our proposed system is the firstwork of leveraging smartphones and IoT environment torecommend tangible travel route within POIs based on onsitetravel behavior sensing and mining methods
3 Research Methodologies
31 System Overview In this work we use the phrase scenicarea to denote a park or a museum containing a series ofsightseeing spots or exhibits namely interesting spots At eachinteresting spot entrance and exit of a scenic area need to bepreinstalled a Bluetooth low energy (BLE) beacon to locatetourists in an indoor or outdoor scenario An illustrativeexample of the system is shown in Figure 1 Concretely oursystem adopts iBeacon [22] devices to indicate specific spotsby broadcasting their own device tags that is positioninginformation When a tourist is approaching an interestingspot with a smart phone the phone will use the locatinginformation to judge whether the tourist has arrived at thisspot If so the phone will record the onsite travel behaviordata at this spot At the end of the travel the phone uploadsa complete behavior sequence and a user-specified profileto the system server Subsequently the server preprocessesthese data transforming them into Tourist-Behavior (TB)pattern sequences and uses the TB pattern mining algo-rithm to generate candidate tangible travel routes At therecommendation stage the system server will recommendpersonalized tangible travel routes for a new tourist by usingthe route ranking method according to the touristrsquos personalprofile and constraintsThe recommended travel route whichcontains a spot visit sequence and their respective visitdurations will help them to finish a valuable tour in the areain a comfortable way
Figure 2 illustrates the workflow of the proposed systemSpecifically stage 1 is performed by a client App running ontouristsrsquo smart phones which is responsible for collecting
4 Wireless Communications and Mobile Computing
EXIT
SPOT G
SPOT D
SPOT C
SPOT F
iBeacon
SPOT B
iBeacon
iBeacon
SPOT A
ENTRANCE
SPOT E
A Client AppSensing Onsite Behavior
iBeacon
System Server
Travel Behavior Data Preprocessing
Behavior Pattern Sequence Mining
Travel RouteRetrievingampRanking
Historical travel behavior of tourist 1Historical travel behavior of tourist 2
iBeacon
iBeacon
iBeacon
iBeacon
iBeacon
Uplo
adin
g Beh
avio
r Se
quen
ces
Input a Personal profile
Recommending a Travel Route
A Scenic Area
A tangible route for a new touristHistorical travel behavior of tourist 3
Figure 1 An illustrative example for the travel route recommendation
Stage3 Tangible travel route recommendation
Travel behavior sequence
preprocessing
The Tourist-Behavior pattern mining
algorithm
Receiving profile and constraints
Retrieving candidate travel routes
Ranking routesby calculating the
rank values
Stage2 Tourist-Behavior mining
Input tourist profiles and constraints
Recommend tangible
travel routes
Sensing onsite travel behavior
Uploading personal profiles and travel
behavior sequences
Stage1 Onsite behavior collecting
Client App System Server Client App
Figure 2 The workflow of the proposed system
touristsrsquo personal profiles and their behavior data whilestages 2 and 3 are performed on the system server sideIn offline running stage 2 behavior sequence preprocessingmethod and Tourist-Behavior (TB) PrefixSpan algorithm areproposed to generate a series of TB pattern sequences that iscandidate tangible travel routes In online running stage 3 thesystem server recommends tangible travel routes for varioustourists based on their profiles and route constraints
32 Onsite Travel Behavior Collecting Since tourists withdifferent personal attributes may have different personalinterests stamina walking speeds and so forth to ensurethe personalization of our recommendations we classifyand store the collected behavior sequences according to
corresponding personal profiles in our system At the begin-ning of the behavior data collection process we request eachtourist to input three common and typical personal attributesincluding gender age group and education level as a simpleprofileThen the client App uploads an onsite travel behaviorsequence and a corresponding profile together to the sys-tem server Subsequently at the recommendation stage thesystem server uses a personal profile of the querying touristto retrieve generated routes from the corresponding routesubset for matching touristsrsquo different interests
321 Positioning Mechanism The positioning mechanismis implemented based on iBeacon devices and smartphones The iBeacon protocol is characterized as low energy
Wireless Communications and Mobile Computing 5
Table 1 A simple example of travel behavior sequences
Sid Onsite behavior sequence01 (lt0Ze300gtlt6A2654gtlt35B4504gtlt55D7022gtlt78F9044gtlt99E10100gtlt108G12834gtlt133Zt13500gt)02 (lt0Ze200gtlt4A2145gtlt31B4012gtlt50C6021gtlt72F8622gtlt92G10613gtlt112Zt11400gt)03 (lt0Ze300gtlt9A2211gtlt34B3800gtlt46C5834gtlt70F8644gtlt90E10864gtlt112G12423gtlt130Zt13200gt)
consumption and broad wireless broadcasting range whichcan be applied in indoor and outdoor scenarios Besidesthere is no pairing connection during the locating processwhich differs from traditional Bluetooth protocolsThereforeiBeacon makes the positioning mechanismmore flexible andefficient
During the tourist locating process the iBeacon devicesconstantly broadcast their own location identities (ID) with aTX power valueThe positioning information consists of two16-bit protocol data fields named major ID and minor IDwhich are used to represent a scenic area and an interestingspot respectively Meanwhile a nearby smartphone adopts(1) to compute a proximity distance d between itself and thebroadcasting iBeacon device to locate itself in a scenic area
119889 = 10and ((|119877119878119878119868| minus 119860)(10 lowast 119899) ) (1)
where119860 is the TX power constant that stands for the receivedsignal strength at 1-meter distance from the iBeacon deviceRSSI is the current BLE signal strength of the smart phone119899 is the path loss coefficient constant and 119889 is a distance ofmeters between the smartphone and the iBeacon device [54]The client App running on a smartphone chooses the lowest119889 as the current recognized interesting spot when the phoneis receiving multiple iBeacon signals simultaneously
322 Travel Behavior Sensing and Recording During theonsite behavior sensing procedure the client App has twotasks (a) reckoning the current interesting spot where thetourist is arriving at meanwhile recording the arriving andleaving timestamps of each interesting spot by comparing thedistance threshold with the real distance between the currentiBeacon device and the smartphone (b) Monitoring the dataof smartphone devices that is the on-board camera andaccelerometer so as to record the behavior of taking picturesand standing still to appreciate something on each interestingspot of the tourist
To record the number of taking pictures behaviorsthe client App monitors the on-board camera operationmessage of Android system namely ldquoandroid hardwareactionNEW PICTURErdquo once the tourist uses the phonecamera to take a picture To record the number of standingbehaviors the client App integrates 3-dimensional acceler-ations into an overall acceleration data first Then it uses aSliding Window Filtering method [55] to count the numberof standing behaviors The client App inserts the number ofthese two behaviors into the current travel behavior sequenceLast the client App uploads the behavior sequence and itscorresponding profile to the system server when it detects theexit of the scenic area
Let 119861 = 1198871 1198872 119887119892 be the set of iBeacon devices thatare installed in a specific scenic area In the system servera travel behavior sequence record is stored as ltsid tbsgtwhere sid is the identifier of the sequence and tbs is an onsitebehavior sequence And tbs consists of a sequence (ltstin1 b1stout1 p1 s1gt ltstin2 b2 stout2 p2 s2gt ltstink bk stoutkpk skgt) where the quintupleltstini bi stouti pi sigt representsa behavior data with respect to the interesting spot i bi is thecorresponding iBeacon device ID and 119887119894 isin 119861 stini and stoutistands for arriving and leaving timestamps respectively andstinilestoutilestini+1 for 1 le 119894 le 119896 minus 1 pi and si are the numberof taking pictures and standing still to appreciate somethingrespectively Further the visit duration of spot i is calculatedby stouti - stini the interval between spot i and spot i+1 iscalculated by 119904119905119894119899119894+1 minus 119904119905119900119906119905119894Example 1 As illustrated in Figure 1 there are one entranceone exit and seven interesting spots in the scenic area Thusthere are nine iBeacon devices as total needed to install inthe area After tourist 4 inputs his or her profile and timeconstraint the system returns a travel route by mining thehistorical travel behavior sequences acquired from the otherthree tourists The corresponding sequences are shown inTable 1 for example tourist 1 visited six interesting spots ABD FE andGThe symbolsZe andZt stand for the entranceand the exit respectively Taking the behavior data at spot Aas an instance tourist 1 arrived at spot A at the 6thmin andleft out at the 26th min took 5 pictures and stood still for 4times at spot A
33 Tourist-Behavior Mining The goal of the Tourist-Behavior mining stage is to generate various candidatetravel routes by mining the historical onsite travel behaviorsequences This stage consists of two steps the travel behav-ior sequence preprocessing step and the Tourist-Behaviorsequential travel routes generating step
331 Travel Behavior Sequence Preprocessing The prepro-cessing step is to transform travel behavior sequences intoTourist-Behavior (TB) pattern sequences and then storepattern sequences into route subset according to their cor-responding personal profile Before describing the details ofthe step the following definitions are given
Definition 2 ATourist-Behavior (TB) pattern 120582119894 is defined asa triple ltbi NPi Di gt where bi is the location identity of spoti NPi is the normalized popularity value about spot i Di isthe discrete visit duration at spot i Note that the pattern 120582119894 issaid to match the pattern 120582119895 if and only if bi = bj NPi = NPjand Di = Dj
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Wireless Communications and Mobile Computing
travel behaviors imply his or her objective preferences andinterests to some objects For instance tourists will spendlonger visit duration or take more pictures or stand stillmore times to appreciate something on a spot if they aremore interested in somethingThus gathering touristsrsquo onsitetravel behaviors and mining their personal preferences andfrequent travel routes could be an effective approach forrecommending tangible travel routes for new similar touristsin specific POIs
To that end in this work we proposed a POI-orientedtravel route recommendation system based on IoT technol-ogy and smart phones In detail we first adopted Bluetoothlow energy (BLE) beacons [22] to periodically broadcastpositioning information for nearby smart phones Also wedeveloped a client App running on an Android smartphoneto collect onsite travel behaviors data and correspondingpersonal profiles and then upload collected data to thesystem server Next on the system server side all collectedtravel behavior data are classified according to their per-sonal profiles And then a behavior sequence preprocessingmethod and Tourist-Behavior pattern mining algorithmswere designed to generate diverse tangible travel routes Atroute recommendation stage to ensure the personalization ofour recommendations the proposed route ranking methodrecommends tangible travel routes for new tourists by usingtheir personal profiles and route constraints As all tangibletravel routes are constructed from real historical onsite travelbehavior the recommended routes have high accuracy andrationality in terms of visit arrangements Also since therecommended routes are retrieved from the correspondingcandidate route subset according to the querying touristrsquo sprofile the visit objects of the final route can suit the personalinterests of the tourist better
Our main contributions are summarized as follows (a)An onsite travel behavior data collecting method which isbased on touristsrsquo smartphones and Bluetooth low energy(BLE) beacons is designed to automatically sense onsitetravel behavior under indoor and outdoor tourism scenarios(b) Tourist-Behavior PrefixSpan algorithm is proposed togenerate diverse frequent travel routes effectively based onhistorical Tourist-Behavior pattern sequences (c) Travelroute ranking method is proposed to recommend a list oftangible travel routes according to the querying touristrsquosprofile and constraints so as to ensure the route value andrationality of the final travel routes (d) Experimental resultsdemonstrate the effectiveness of our system in recommend-ing personalized tangible travel routes for tourists in a givenPOI based on historical onsite travel behavior
The rest of the paper is organized as follows Section 2discusses the related works regarding travel route recom-mendation systems and tourism recommendationwithin IoTenvironment Section 3 presents the research methodologiesof the proposed system where the framework of the system isdescribed in Section 31 onsite travel behavior data collectingmethod is explained in Section 32 The Tourist-Behaviorpattern sequences mining method and tangible travel routerecommendation procedure are thoroughly explained in Sec-tions 33 and 34 respectively Section 4 analyzes the experi-mental results to validate the feasibility and performance of
the proposed system The conclusion and future works arepresented in Section 5
2 Related Works
21 Travel Route Recommendation Systems Due to the prac-tical values of travel route recommendation systems lots ofresearchers have been placing a great emphasis on solvingthe route planning in tourism scenarios [19 23] in recentyears One category of these works uses the Orienteeringproblem [24] and its variants to approach the route planningproblem These methods formulate the problem from differ-ent perspectives resulting in diverse problem models whichconsider different problem variables and constraints Theroute-generation process actually is a near-optimal solutionusing metaheuristic searching algorithm [25] Accordinglyto enhance the personalization of the recommended routesmost of these works resort to acquiring more detailed userfeedbacks or profiles to assist in fine-tuning the final resultsIn [26] the system solicits walking travel related attributesfrom tourists to insert concrete walking routes into POIitineraries thereby supporting more experiential explorationof tourist destinations Zhang et al [27 28] studied tourrecommendation with the goal of recommending person-alized itineraries based on the interest preferences of usersand available touring time while considering opening hoursof POIs and uncertainty in travelling time Other studiesconsider more practical factors that raise novel optimizationchallenges incorporating forms of situational awareness suchas multiple modes of transport [29] considering trafficconditions [30ndash32] POI crowdedness [33 34] and queuingtimes [35]
Although the optimization-based route planning systemscan recommend a reasonable travel route adhering to onersquospreferences and constraints the interactive preferences inputof the planning process is time costly to tourists It isimpractical for tourists to spend a long time in inputtinga complex user profile when entering a specific POI Fur-thermore the results of these systems are lack of diversityand less personalization due to the near-optimal solutionsearching methodology Therefore lots of other works focuson generating personalized travel routes by mining UserGenerated Contents (UGC) that is data-driven approachesto route planning The UGC adopted in previous researchesinclude GPS trajectory datasets [36] check-in datasets [3738] and geo-tagged photos [39]
Chen et al [31] adopted historical check-in data and GPStrajectories to construct a POI network and used a heuristicmethod to generate a favorite POIs list for a specific user in aninteractive manner Subsequently the system requests usersto specify their favorite POIs during the route-generatingstage PersTour system [13 40] uses geo-tagged photos todetermine POI and construct POI travel sequences andleverage an Orienteering problem solving model to recom-mend POI itineraries by both considering POI popularityand tourist personal interests Majid et al [41] inferred thelocation of POIs and their semantic meaning using clusteringapproaches on geo-tagged photos and used a pattern mining
Wireless Communications and Mobile Computing 3
algorithm to discover popular travel sequences under thecontext of the tour recommendation that is time day andweather Besides the rapid growth of online tourismwebsitesprovides massive POI reviews and travelogues Thus somerecent works [14 42ndash44] focused on generating personalizedmining travelogues and POI related contents
The data-driven based route planning systems can rec-ommend rather personalized and reasonable travel routeshowever the limitation of these systems is that they aimat recommending city-level or district-level orienting POIsitineraries that is planning POIs travel routes for touristswithin a city or region They fail to generate tangible travelroutes for tourists within a specific POI due to lack ofrich onsite travel behavior and related itinerary miningalgorithms
22 Personalized Recommendation in IoT Environment TheIoT concept was first coined by Kevin Ashton in 1999 [45]in supply chain management applications based on radiofrequency identification devices (abbr RFID) At present IoTis referring to a bundle of technologies that aim at sensinghandling and transmitting state information of physicalenvironments which is broadly applied in smart cities [4647] smart business [48 49] and smart tourism scenarios[50]The goal of these smart systems is to recommend a set ofpersonalized and valuable items or services for various usersTo this end researchers focus on recording and analyzinguser behavior to learn user preferences more precisely bydeploying IoT technologies
Specifically in smart tourism applications some studiesfocus on using IoT technologies and mobile devices toimprove tourism experiences in an interactive way Kuusiket al [51] designed a smart museum system that integratesPDAs and RFID technologies to provide users with culturalcontents by sensing the interactive behavior between PDAsand RFID tags which were installed near each artworkIn [52] an indoor location-aware system was designed fora smart museum to enhance visitorsrsquo cultural experiencesThe proposed system obtains visitors localization informa-tion through a Bluetooth low energy (BLE) infrastructureinstalled in the museum and uses several location-aware ser-vices hosted in the system to interact with visitors accordingto their locations
And some other works aim at solving the next visitingspot recommendation problem within a specific POI Mas-simo et al [5] leveraged Inverse Reinforcement Learningmethod to learning user preferences by observing touristsonsite behavior in an IoT-equipped smart museum so as topredict next exhibit sequentially for tourists Hashemi et al[6 7] solved the challenging next POI recommending prob-lem by logging and mining usersrsquo onsite physical and onlineinteraction behavior data within an IoT-augmentedmuseumHowever the above works fail to generate personalized andtangible travel routes for tourists
To that end some researchers strived to solve thischallenging problem by mining historical travel trajectoriesin an IoT-augmented environment Tsai et al [15] adoptedRFID infrastructure to record visitorsrsquo check-in sequencesof recreation facilities in a theme park and then proposed a
statistical method to find behavioral similar historical visitorsso as to suggest a travel route for the current queryingvisitor Luo et al [16] studied a new path finding system thatdiscovers the most frequent path during user-specified timeperiods in large-scale historical trajectory data Tsai et al[17] proposed a touring path suggesting system for visitorsto comprehend exhibits in exhibitions or museums Thesystem takes previous popular visiting trajectories as the sug-gestion foundation and provides a time-interval sequentialpatterns mining algorithm improved from [18] to generatepersonalized travel routes However as the above systemsonly resorted to dedicated IoT devices to record the check-in behavior they hardly learn more tangible user preferencestowards each interest object from the single dimensionalbehavior Meanwhile current smartphones generally equipa camera and diverse sensors which could be used tosense multiple dimensional onsite behaviors of tourists soas to explore high-level tourist preferences and recommendpersonalized tangible travel route Although some previousresearches have investigated the human activity recognitionbased on smartphone sensors [21 53] there is no study onlearning user preferences directly by smartphones To thebest of our knowledge our proposed system is the firstwork of leveraging smartphones and IoT environment torecommend tangible travel route within POIs based on onsitetravel behavior sensing and mining methods
3 Research Methodologies
31 System Overview In this work we use the phrase scenicarea to denote a park or a museum containing a series ofsightseeing spots or exhibits namely interesting spots At eachinteresting spot entrance and exit of a scenic area need to bepreinstalled a Bluetooth low energy (BLE) beacon to locatetourists in an indoor or outdoor scenario An illustrativeexample of the system is shown in Figure 1 Concretely oursystem adopts iBeacon [22] devices to indicate specific spotsby broadcasting their own device tags that is positioninginformation When a tourist is approaching an interestingspot with a smart phone the phone will use the locatinginformation to judge whether the tourist has arrived at thisspot If so the phone will record the onsite travel behaviordata at this spot At the end of the travel the phone uploadsa complete behavior sequence and a user-specified profileto the system server Subsequently the server preprocessesthese data transforming them into Tourist-Behavior (TB)pattern sequences and uses the TB pattern mining algo-rithm to generate candidate tangible travel routes At therecommendation stage the system server will recommendpersonalized tangible travel routes for a new tourist by usingthe route ranking method according to the touristrsquos personalprofile and constraintsThe recommended travel route whichcontains a spot visit sequence and their respective visitdurations will help them to finish a valuable tour in the areain a comfortable way
Figure 2 illustrates the workflow of the proposed systemSpecifically stage 1 is performed by a client App running ontouristsrsquo smart phones which is responsible for collecting
4 Wireless Communications and Mobile Computing
EXIT
SPOT G
SPOT D
SPOT C
SPOT F
iBeacon
SPOT B
iBeacon
iBeacon
SPOT A
ENTRANCE
SPOT E
A Client AppSensing Onsite Behavior
iBeacon
System Server
Travel Behavior Data Preprocessing
Behavior Pattern Sequence Mining
Travel RouteRetrievingampRanking
Historical travel behavior of tourist 1Historical travel behavior of tourist 2
iBeacon
iBeacon
iBeacon
iBeacon
iBeacon
Uplo
adin
g Beh
avio
r Se
quen
ces
Input a Personal profile
Recommending a Travel Route
A Scenic Area
A tangible route for a new touristHistorical travel behavior of tourist 3
Figure 1 An illustrative example for the travel route recommendation
Stage3 Tangible travel route recommendation
Travel behavior sequence
preprocessing
The Tourist-Behavior pattern mining
algorithm
Receiving profile and constraints
Retrieving candidate travel routes
Ranking routesby calculating the
rank values
Stage2 Tourist-Behavior mining
Input tourist profiles and constraints
Recommend tangible
travel routes
Sensing onsite travel behavior
Uploading personal profiles and travel
behavior sequences
Stage1 Onsite behavior collecting
Client App System Server Client App
Figure 2 The workflow of the proposed system
touristsrsquo personal profiles and their behavior data whilestages 2 and 3 are performed on the system server sideIn offline running stage 2 behavior sequence preprocessingmethod and Tourist-Behavior (TB) PrefixSpan algorithm areproposed to generate a series of TB pattern sequences that iscandidate tangible travel routes In online running stage 3 thesystem server recommends tangible travel routes for varioustourists based on their profiles and route constraints
32 Onsite Travel Behavior Collecting Since tourists withdifferent personal attributes may have different personalinterests stamina walking speeds and so forth to ensurethe personalization of our recommendations we classifyand store the collected behavior sequences according to
corresponding personal profiles in our system At the begin-ning of the behavior data collection process we request eachtourist to input three common and typical personal attributesincluding gender age group and education level as a simpleprofileThen the client App uploads an onsite travel behaviorsequence and a corresponding profile together to the sys-tem server Subsequently at the recommendation stage thesystem server uses a personal profile of the querying touristto retrieve generated routes from the corresponding routesubset for matching touristsrsquo different interests
321 Positioning Mechanism The positioning mechanismis implemented based on iBeacon devices and smartphones The iBeacon protocol is characterized as low energy
Wireless Communications and Mobile Computing 5
Table 1 A simple example of travel behavior sequences
Sid Onsite behavior sequence01 (lt0Ze300gtlt6A2654gtlt35B4504gtlt55D7022gtlt78F9044gtlt99E10100gtlt108G12834gtlt133Zt13500gt)02 (lt0Ze200gtlt4A2145gtlt31B4012gtlt50C6021gtlt72F8622gtlt92G10613gtlt112Zt11400gt)03 (lt0Ze300gtlt9A2211gtlt34B3800gtlt46C5834gtlt70F8644gtlt90E10864gtlt112G12423gtlt130Zt13200gt)
consumption and broad wireless broadcasting range whichcan be applied in indoor and outdoor scenarios Besidesthere is no pairing connection during the locating processwhich differs from traditional Bluetooth protocolsThereforeiBeacon makes the positioning mechanismmore flexible andefficient
During the tourist locating process the iBeacon devicesconstantly broadcast their own location identities (ID) with aTX power valueThe positioning information consists of two16-bit protocol data fields named major ID and minor IDwhich are used to represent a scenic area and an interestingspot respectively Meanwhile a nearby smartphone adopts(1) to compute a proximity distance d between itself and thebroadcasting iBeacon device to locate itself in a scenic area
119889 = 10and ((|119877119878119878119868| minus 119860)(10 lowast 119899) ) (1)
where119860 is the TX power constant that stands for the receivedsignal strength at 1-meter distance from the iBeacon deviceRSSI is the current BLE signal strength of the smart phone119899 is the path loss coefficient constant and 119889 is a distance ofmeters between the smartphone and the iBeacon device [54]The client App running on a smartphone chooses the lowest119889 as the current recognized interesting spot when the phoneis receiving multiple iBeacon signals simultaneously
322 Travel Behavior Sensing and Recording During theonsite behavior sensing procedure the client App has twotasks (a) reckoning the current interesting spot where thetourist is arriving at meanwhile recording the arriving andleaving timestamps of each interesting spot by comparing thedistance threshold with the real distance between the currentiBeacon device and the smartphone (b) Monitoring the dataof smartphone devices that is the on-board camera andaccelerometer so as to record the behavior of taking picturesand standing still to appreciate something on each interestingspot of the tourist
To record the number of taking pictures behaviorsthe client App monitors the on-board camera operationmessage of Android system namely ldquoandroid hardwareactionNEW PICTURErdquo once the tourist uses the phonecamera to take a picture To record the number of standingbehaviors the client App integrates 3-dimensional acceler-ations into an overall acceleration data first Then it uses aSliding Window Filtering method [55] to count the numberof standing behaviors The client App inserts the number ofthese two behaviors into the current travel behavior sequenceLast the client App uploads the behavior sequence and itscorresponding profile to the system server when it detects theexit of the scenic area
Let 119861 = 1198871 1198872 119887119892 be the set of iBeacon devices thatare installed in a specific scenic area In the system servera travel behavior sequence record is stored as ltsid tbsgtwhere sid is the identifier of the sequence and tbs is an onsitebehavior sequence And tbs consists of a sequence (ltstin1 b1stout1 p1 s1gt ltstin2 b2 stout2 p2 s2gt ltstink bk stoutkpk skgt) where the quintupleltstini bi stouti pi sigt representsa behavior data with respect to the interesting spot i bi is thecorresponding iBeacon device ID and 119887119894 isin 119861 stini and stoutistands for arriving and leaving timestamps respectively andstinilestoutilestini+1 for 1 le 119894 le 119896 minus 1 pi and si are the numberof taking pictures and standing still to appreciate somethingrespectively Further the visit duration of spot i is calculatedby stouti - stini the interval between spot i and spot i+1 iscalculated by 119904119905119894119899119894+1 minus 119904119905119900119906119905119894Example 1 As illustrated in Figure 1 there are one entranceone exit and seven interesting spots in the scenic area Thusthere are nine iBeacon devices as total needed to install inthe area After tourist 4 inputs his or her profile and timeconstraint the system returns a travel route by mining thehistorical travel behavior sequences acquired from the otherthree tourists The corresponding sequences are shown inTable 1 for example tourist 1 visited six interesting spots ABD FE andGThe symbolsZe andZt stand for the entranceand the exit respectively Taking the behavior data at spot Aas an instance tourist 1 arrived at spot A at the 6thmin andleft out at the 26th min took 5 pictures and stood still for 4times at spot A
33 Tourist-Behavior Mining The goal of the Tourist-Behavior mining stage is to generate various candidatetravel routes by mining the historical onsite travel behaviorsequences This stage consists of two steps the travel behav-ior sequence preprocessing step and the Tourist-Behaviorsequential travel routes generating step
331 Travel Behavior Sequence Preprocessing The prepro-cessing step is to transform travel behavior sequences intoTourist-Behavior (TB) pattern sequences and then storepattern sequences into route subset according to their cor-responding personal profile Before describing the details ofthe step the following definitions are given
Definition 2 ATourist-Behavior (TB) pattern 120582119894 is defined asa triple ltbi NPi Di gt where bi is the location identity of spoti NPi is the normalized popularity value about spot i Di isthe discrete visit duration at spot i Note that the pattern 120582119894 issaid to match the pattern 120582119895 if and only if bi = bj NPi = NPjand Di = Dj
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 3
algorithm to discover popular travel sequences under thecontext of the tour recommendation that is time day andweather Besides the rapid growth of online tourismwebsitesprovides massive POI reviews and travelogues Thus somerecent works [14 42ndash44] focused on generating personalizedmining travelogues and POI related contents
The data-driven based route planning systems can rec-ommend rather personalized and reasonable travel routeshowever the limitation of these systems is that they aimat recommending city-level or district-level orienting POIsitineraries that is planning POIs travel routes for touristswithin a city or region They fail to generate tangible travelroutes for tourists within a specific POI due to lack ofrich onsite travel behavior and related itinerary miningalgorithms
22 Personalized Recommendation in IoT Environment TheIoT concept was first coined by Kevin Ashton in 1999 [45]in supply chain management applications based on radiofrequency identification devices (abbr RFID) At present IoTis referring to a bundle of technologies that aim at sensinghandling and transmitting state information of physicalenvironments which is broadly applied in smart cities [4647] smart business [48 49] and smart tourism scenarios[50]The goal of these smart systems is to recommend a set ofpersonalized and valuable items or services for various usersTo this end researchers focus on recording and analyzinguser behavior to learn user preferences more precisely bydeploying IoT technologies
Specifically in smart tourism applications some studiesfocus on using IoT technologies and mobile devices toimprove tourism experiences in an interactive way Kuusiket al [51] designed a smart museum system that integratesPDAs and RFID technologies to provide users with culturalcontents by sensing the interactive behavior between PDAsand RFID tags which were installed near each artworkIn [52] an indoor location-aware system was designed fora smart museum to enhance visitorsrsquo cultural experiencesThe proposed system obtains visitors localization informa-tion through a Bluetooth low energy (BLE) infrastructureinstalled in the museum and uses several location-aware ser-vices hosted in the system to interact with visitors accordingto their locations
And some other works aim at solving the next visitingspot recommendation problem within a specific POI Mas-simo et al [5] leveraged Inverse Reinforcement Learningmethod to learning user preferences by observing touristsonsite behavior in an IoT-equipped smart museum so as topredict next exhibit sequentially for tourists Hashemi et al[6 7] solved the challenging next POI recommending prob-lem by logging and mining usersrsquo onsite physical and onlineinteraction behavior data within an IoT-augmentedmuseumHowever the above works fail to generate personalized andtangible travel routes for tourists
To that end some researchers strived to solve thischallenging problem by mining historical travel trajectoriesin an IoT-augmented environment Tsai et al [15] adoptedRFID infrastructure to record visitorsrsquo check-in sequencesof recreation facilities in a theme park and then proposed a
statistical method to find behavioral similar historical visitorsso as to suggest a travel route for the current queryingvisitor Luo et al [16] studied a new path finding system thatdiscovers the most frequent path during user-specified timeperiods in large-scale historical trajectory data Tsai et al[17] proposed a touring path suggesting system for visitorsto comprehend exhibits in exhibitions or museums Thesystem takes previous popular visiting trajectories as the sug-gestion foundation and provides a time-interval sequentialpatterns mining algorithm improved from [18] to generatepersonalized travel routes However as the above systemsonly resorted to dedicated IoT devices to record the check-in behavior they hardly learn more tangible user preferencestowards each interest object from the single dimensionalbehavior Meanwhile current smartphones generally equipa camera and diverse sensors which could be used tosense multiple dimensional onsite behaviors of tourists soas to explore high-level tourist preferences and recommendpersonalized tangible travel route Although some previousresearches have investigated the human activity recognitionbased on smartphone sensors [21 53] there is no study onlearning user preferences directly by smartphones To thebest of our knowledge our proposed system is the firstwork of leveraging smartphones and IoT environment torecommend tangible travel route within POIs based on onsitetravel behavior sensing and mining methods
3 Research Methodologies
31 System Overview In this work we use the phrase scenicarea to denote a park or a museum containing a series ofsightseeing spots or exhibits namely interesting spots At eachinteresting spot entrance and exit of a scenic area need to bepreinstalled a Bluetooth low energy (BLE) beacon to locatetourists in an indoor or outdoor scenario An illustrativeexample of the system is shown in Figure 1 Concretely oursystem adopts iBeacon [22] devices to indicate specific spotsby broadcasting their own device tags that is positioninginformation When a tourist is approaching an interestingspot with a smart phone the phone will use the locatinginformation to judge whether the tourist has arrived at thisspot If so the phone will record the onsite travel behaviordata at this spot At the end of the travel the phone uploadsa complete behavior sequence and a user-specified profileto the system server Subsequently the server preprocessesthese data transforming them into Tourist-Behavior (TB)pattern sequences and uses the TB pattern mining algo-rithm to generate candidate tangible travel routes At therecommendation stage the system server will recommendpersonalized tangible travel routes for a new tourist by usingthe route ranking method according to the touristrsquos personalprofile and constraintsThe recommended travel route whichcontains a spot visit sequence and their respective visitdurations will help them to finish a valuable tour in the areain a comfortable way
Figure 2 illustrates the workflow of the proposed systemSpecifically stage 1 is performed by a client App running ontouristsrsquo smart phones which is responsible for collecting
4 Wireless Communications and Mobile Computing
EXIT
SPOT G
SPOT D
SPOT C
SPOT F
iBeacon
SPOT B
iBeacon
iBeacon
SPOT A
ENTRANCE
SPOT E
A Client AppSensing Onsite Behavior
iBeacon
System Server
Travel Behavior Data Preprocessing
Behavior Pattern Sequence Mining
Travel RouteRetrievingampRanking
Historical travel behavior of tourist 1Historical travel behavior of tourist 2
iBeacon
iBeacon
iBeacon
iBeacon
iBeacon
Uplo
adin
g Beh
avio
r Se
quen
ces
Input a Personal profile
Recommending a Travel Route
A Scenic Area
A tangible route for a new touristHistorical travel behavior of tourist 3
Figure 1 An illustrative example for the travel route recommendation
Stage3 Tangible travel route recommendation
Travel behavior sequence
preprocessing
The Tourist-Behavior pattern mining
algorithm
Receiving profile and constraints
Retrieving candidate travel routes
Ranking routesby calculating the
rank values
Stage2 Tourist-Behavior mining
Input tourist profiles and constraints
Recommend tangible
travel routes
Sensing onsite travel behavior
Uploading personal profiles and travel
behavior sequences
Stage1 Onsite behavior collecting
Client App System Server Client App
Figure 2 The workflow of the proposed system
touristsrsquo personal profiles and their behavior data whilestages 2 and 3 are performed on the system server sideIn offline running stage 2 behavior sequence preprocessingmethod and Tourist-Behavior (TB) PrefixSpan algorithm areproposed to generate a series of TB pattern sequences that iscandidate tangible travel routes In online running stage 3 thesystem server recommends tangible travel routes for varioustourists based on their profiles and route constraints
32 Onsite Travel Behavior Collecting Since tourists withdifferent personal attributes may have different personalinterests stamina walking speeds and so forth to ensurethe personalization of our recommendations we classifyand store the collected behavior sequences according to
corresponding personal profiles in our system At the begin-ning of the behavior data collection process we request eachtourist to input three common and typical personal attributesincluding gender age group and education level as a simpleprofileThen the client App uploads an onsite travel behaviorsequence and a corresponding profile together to the sys-tem server Subsequently at the recommendation stage thesystem server uses a personal profile of the querying touristto retrieve generated routes from the corresponding routesubset for matching touristsrsquo different interests
321 Positioning Mechanism The positioning mechanismis implemented based on iBeacon devices and smartphones The iBeacon protocol is characterized as low energy
Wireless Communications and Mobile Computing 5
Table 1 A simple example of travel behavior sequences
Sid Onsite behavior sequence01 (lt0Ze300gtlt6A2654gtlt35B4504gtlt55D7022gtlt78F9044gtlt99E10100gtlt108G12834gtlt133Zt13500gt)02 (lt0Ze200gtlt4A2145gtlt31B4012gtlt50C6021gtlt72F8622gtlt92G10613gtlt112Zt11400gt)03 (lt0Ze300gtlt9A2211gtlt34B3800gtlt46C5834gtlt70F8644gtlt90E10864gtlt112G12423gtlt130Zt13200gt)
consumption and broad wireless broadcasting range whichcan be applied in indoor and outdoor scenarios Besidesthere is no pairing connection during the locating processwhich differs from traditional Bluetooth protocolsThereforeiBeacon makes the positioning mechanismmore flexible andefficient
During the tourist locating process the iBeacon devicesconstantly broadcast their own location identities (ID) with aTX power valueThe positioning information consists of two16-bit protocol data fields named major ID and minor IDwhich are used to represent a scenic area and an interestingspot respectively Meanwhile a nearby smartphone adopts(1) to compute a proximity distance d between itself and thebroadcasting iBeacon device to locate itself in a scenic area
119889 = 10and ((|119877119878119878119868| minus 119860)(10 lowast 119899) ) (1)
where119860 is the TX power constant that stands for the receivedsignal strength at 1-meter distance from the iBeacon deviceRSSI is the current BLE signal strength of the smart phone119899 is the path loss coefficient constant and 119889 is a distance ofmeters between the smartphone and the iBeacon device [54]The client App running on a smartphone chooses the lowest119889 as the current recognized interesting spot when the phoneis receiving multiple iBeacon signals simultaneously
322 Travel Behavior Sensing and Recording During theonsite behavior sensing procedure the client App has twotasks (a) reckoning the current interesting spot where thetourist is arriving at meanwhile recording the arriving andleaving timestamps of each interesting spot by comparing thedistance threshold with the real distance between the currentiBeacon device and the smartphone (b) Monitoring the dataof smartphone devices that is the on-board camera andaccelerometer so as to record the behavior of taking picturesand standing still to appreciate something on each interestingspot of the tourist
To record the number of taking pictures behaviorsthe client App monitors the on-board camera operationmessage of Android system namely ldquoandroid hardwareactionNEW PICTURErdquo once the tourist uses the phonecamera to take a picture To record the number of standingbehaviors the client App integrates 3-dimensional acceler-ations into an overall acceleration data first Then it uses aSliding Window Filtering method [55] to count the numberof standing behaviors The client App inserts the number ofthese two behaviors into the current travel behavior sequenceLast the client App uploads the behavior sequence and itscorresponding profile to the system server when it detects theexit of the scenic area
Let 119861 = 1198871 1198872 119887119892 be the set of iBeacon devices thatare installed in a specific scenic area In the system servera travel behavior sequence record is stored as ltsid tbsgtwhere sid is the identifier of the sequence and tbs is an onsitebehavior sequence And tbs consists of a sequence (ltstin1 b1stout1 p1 s1gt ltstin2 b2 stout2 p2 s2gt ltstink bk stoutkpk skgt) where the quintupleltstini bi stouti pi sigt representsa behavior data with respect to the interesting spot i bi is thecorresponding iBeacon device ID and 119887119894 isin 119861 stini and stoutistands for arriving and leaving timestamps respectively andstinilestoutilestini+1 for 1 le 119894 le 119896 minus 1 pi and si are the numberof taking pictures and standing still to appreciate somethingrespectively Further the visit duration of spot i is calculatedby stouti - stini the interval between spot i and spot i+1 iscalculated by 119904119905119894119899119894+1 minus 119904119905119900119906119905119894Example 1 As illustrated in Figure 1 there are one entranceone exit and seven interesting spots in the scenic area Thusthere are nine iBeacon devices as total needed to install inthe area After tourist 4 inputs his or her profile and timeconstraint the system returns a travel route by mining thehistorical travel behavior sequences acquired from the otherthree tourists The corresponding sequences are shown inTable 1 for example tourist 1 visited six interesting spots ABD FE andGThe symbolsZe andZt stand for the entranceand the exit respectively Taking the behavior data at spot Aas an instance tourist 1 arrived at spot A at the 6thmin andleft out at the 26th min took 5 pictures and stood still for 4times at spot A
33 Tourist-Behavior Mining The goal of the Tourist-Behavior mining stage is to generate various candidatetravel routes by mining the historical onsite travel behaviorsequences This stage consists of two steps the travel behav-ior sequence preprocessing step and the Tourist-Behaviorsequential travel routes generating step
331 Travel Behavior Sequence Preprocessing The prepro-cessing step is to transform travel behavior sequences intoTourist-Behavior (TB) pattern sequences and then storepattern sequences into route subset according to their cor-responding personal profile Before describing the details ofthe step the following definitions are given
Definition 2 ATourist-Behavior (TB) pattern 120582119894 is defined asa triple ltbi NPi Di gt where bi is the location identity of spoti NPi is the normalized popularity value about spot i Di isthe discrete visit duration at spot i Note that the pattern 120582119894 issaid to match the pattern 120582119895 if and only if bi = bj NPi = NPjand Di = Dj
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
4 Wireless Communications and Mobile Computing
EXIT
SPOT G
SPOT D
SPOT C
SPOT F
iBeacon
SPOT B
iBeacon
iBeacon
SPOT A
ENTRANCE
SPOT E
A Client AppSensing Onsite Behavior
iBeacon
System Server
Travel Behavior Data Preprocessing
Behavior Pattern Sequence Mining
Travel RouteRetrievingampRanking
Historical travel behavior of tourist 1Historical travel behavior of tourist 2
iBeacon
iBeacon
iBeacon
iBeacon
iBeacon
Uplo
adin
g Beh
avio
r Se
quen
ces
Input a Personal profile
Recommending a Travel Route
A Scenic Area
A tangible route for a new touristHistorical travel behavior of tourist 3
Figure 1 An illustrative example for the travel route recommendation
Stage3 Tangible travel route recommendation
Travel behavior sequence
preprocessing
The Tourist-Behavior pattern mining
algorithm
Receiving profile and constraints
Retrieving candidate travel routes
Ranking routesby calculating the
rank values
Stage2 Tourist-Behavior mining
Input tourist profiles and constraints
Recommend tangible
travel routes
Sensing onsite travel behavior
Uploading personal profiles and travel
behavior sequences
Stage1 Onsite behavior collecting
Client App System Server Client App
Figure 2 The workflow of the proposed system
touristsrsquo personal profiles and their behavior data whilestages 2 and 3 are performed on the system server sideIn offline running stage 2 behavior sequence preprocessingmethod and Tourist-Behavior (TB) PrefixSpan algorithm areproposed to generate a series of TB pattern sequences that iscandidate tangible travel routes In online running stage 3 thesystem server recommends tangible travel routes for varioustourists based on their profiles and route constraints
32 Onsite Travel Behavior Collecting Since tourists withdifferent personal attributes may have different personalinterests stamina walking speeds and so forth to ensurethe personalization of our recommendations we classifyand store the collected behavior sequences according to
corresponding personal profiles in our system At the begin-ning of the behavior data collection process we request eachtourist to input three common and typical personal attributesincluding gender age group and education level as a simpleprofileThen the client App uploads an onsite travel behaviorsequence and a corresponding profile together to the sys-tem server Subsequently at the recommendation stage thesystem server uses a personal profile of the querying touristto retrieve generated routes from the corresponding routesubset for matching touristsrsquo different interests
321 Positioning Mechanism The positioning mechanismis implemented based on iBeacon devices and smartphones The iBeacon protocol is characterized as low energy
Wireless Communications and Mobile Computing 5
Table 1 A simple example of travel behavior sequences
Sid Onsite behavior sequence01 (lt0Ze300gtlt6A2654gtlt35B4504gtlt55D7022gtlt78F9044gtlt99E10100gtlt108G12834gtlt133Zt13500gt)02 (lt0Ze200gtlt4A2145gtlt31B4012gtlt50C6021gtlt72F8622gtlt92G10613gtlt112Zt11400gt)03 (lt0Ze300gtlt9A2211gtlt34B3800gtlt46C5834gtlt70F8644gtlt90E10864gtlt112G12423gtlt130Zt13200gt)
consumption and broad wireless broadcasting range whichcan be applied in indoor and outdoor scenarios Besidesthere is no pairing connection during the locating processwhich differs from traditional Bluetooth protocolsThereforeiBeacon makes the positioning mechanismmore flexible andefficient
During the tourist locating process the iBeacon devicesconstantly broadcast their own location identities (ID) with aTX power valueThe positioning information consists of two16-bit protocol data fields named major ID and minor IDwhich are used to represent a scenic area and an interestingspot respectively Meanwhile a nearby smartphone adopts(1) to compute a proximity distance d between itself and thebroadcasting iBeacon device to locate itself in a scenic area
119889 = 10and ((|119877119878119878119868| minus 119860)(10 lowast 119899) ) (1)
where119860 is the TX power constant that stands for the receivedsignal strength at 1-meter distance from the iBeacon deviceRSSI is the current BLE signal strength of the smart phone119899 is the path loss coefficient constant and 119889 is a distance ofmeters between the smartphone and the iBeacon device [54]The client App running on a smartphone chooses the lowest119889 as the current recognized interesting spot when the phoneis receiving multiple iBeacon signals simultaneously
322 Travel Behavior Sensing and Recording During theonsite behavior sensing procedure the client App has twotasks (a) reckoning the current interesting spot where thetourist is arriving at meanwhile recording the arriving andleaving timestamps of each interesting spot by comparing thedistance threshold with the real distance between the currentiBeacon device and the smartphone (b) Monitoring the dataof smartphone devices that is the on-board camera andaccelerometer so as to record the behavior of taking picturesand standing still to appreciate something on each interestingspot of the tourist
To record the number of taking pictures behaviorsthe client App monitors the on-board camera operationmessage of Android system namely ldquoandroid hardwareactionNEW PICTURErdquo once the tourist uses the phonecamera to take a picture To record the number of standingbehaviors the client App integrates 3-dimensional acceler-ations into an overall acceleration data first Then it uses aSliding Window Filtering method [55] to count the numberof standing behaviors The client App inserts the number ofthese two behaviors into the current travel behavior sequenceLast the client App uploads the behavior sequence and itscorresponding profile to the system server when it detects theexit of the scenic area
Let 119861 = 1198871 1198872 119887119892 be the set of iBeacon devices thatare installed in a specific scenic area In the system servera travel behavior sequence record is stored as ltsid tbsgtwhere sid is the identifier of the sequence and tbs is an onsitebehavior sequence And tbs consists of a sequence (ltstin1 b1stout1 p1 s1gt ltstin2 b2 stout2 p2 s2gt ltstink bk stoutkpk skgt) where the quintupleltstini bi stouti pi sigt representsa behavior data with respect to the interesting spot i bi is thecorresponding iBeacon device ID and 119887119894 isin 119861 stini and stoutistands for arriving and leaving timestamps respectively andstinilestoutilestini+1 for 1 le 119894 le 119896 minus 1 pi and si are the numberof taking pictures and standing still to appreciate somethingrespectively Further the visit duration of spot i is calculatedby stouti - stini the interval between spot i and spot i+1 iscalculated by 119904119905119894119899119894+1 minus 119904119905119900119906119905119894Example 1 As illustrated in Figure 1 there are one entranceone exit and seven interesting spots in the scenic area Thusthere are nine iBeacon devices as total needed to install inthe area After tourist 4 inputs his or her profile and timeconstraint the system returns a travel route by mining thehistorical travel behavior sequences acquired from the otherthree tourists The corresponding sequences are shown inTable 1 for example tourist 1 visited six interesting spots ABD FE andGThe symbolsZe andZt stand for the entranceand the exit respectively Taking the behavior data at spot Aas an instance tourist 1 arrived at spot A at the 6thmin andleft out at the 26th min took 5 pictures and stood still for 4times at spot A
33 Tourist-Behavior Mining The goal of the Tourist-Behavior mining stage is to generate various candidatetravel routes by mining the historical onsite travel behaviorsequences This stage consists of two steps the travel behav-ior sequence preprocessing step and the Tourist-Behaviorsequential travel routes generating step
331 Travel Behavior Sequence Preprocessing The prepro-cessing step is to transform travel behavior sequences intoTourist-Behavior (TB) pattern sequences and then storepattern sequences into route subset according to their cor-responding personal profile Before describing the details ofthe step the following definitions are given
Definition 2 ATourist-Behavior (TB) pattern 120582119894 is defined asa triple ltbi NPi Di gt where bi is the location identity of spoti NPi is the normalized popularity value about spot i Di isthe discrete visit duration at spot i Note that the pattern 120582119894 issaid to match the pattern 120582119895 if and only if bi = bj NPi = NPjand Di = Dj
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 5
Table 1 A simple example of travel behavior sequences
Sid Onsite behavior sequence01 (lt0Ze300gtlt6A2654gtlt35B4504gtlt55D7022gtlt78F9044gtlt99E10100gtlt108G12834gtlt133Zt13500gt)02 (lt0Ze200gtlt4A2145gtlt31B4012gtlt50C6021gtlt72F8622gtlt92G10613gtlt112Zt11400gt)03 (lt0Ze300gtlt9A2211gtlt34B3800gtlt46C5834gtlt70F8644gtlt90E10864gtlt112G12423gtlt130Zt13200gt)
consumption and broad wireless broadcasting range whichcan be applied in indoor and outdoor scenarios Besidesthere is no pairing connection during the locating processwhich differs from traditional Bluetooth protocolsThereforeiBeacon makes the positioning mechanismmore flexible andefficient
During the tourist locating process the iBeacon devicesconstantly broadcast their own location identities (ID) with aTX power valueThe positioning information consists of two16-bit protocol data fields named major ID and minor IDwhich are used to represent a scenic area and an interestingspot respectively Meanwhile a nearby smartphone adopts(1) to compute a proximity distance d between itself and thebroadcasting iBeacon device to locate itself in a scenic area
119889 = 10and ((|119877119878119878119868| minus 119860)(10 lowast 119899) ) (1)
where119860 is the TX power constant that stands for the receivedsignal strength at 1-meter distance from the iBeacon deviceRSSI is the current BLE signal strength of the smart phone119899 is the path loss coefficient constant and 119889 is a distance ofmeters between the smartphone and the iBeacon device [54]The client App running on a smartphone chooses the lowest119889 as the current recognized interesting spot when the phoneis receiving multiple iBeacon signals simultaneously
322 Travel Behavior Sensing and Recording During theonsite behavior sensing procedure the client App has twotasks (a) reckoning the current interesting spot where thetourist is arriving at meanwhile recording the arriving andleaving timestamps of each interesting spot by comparing thedistance threshold with the real distance between the currentiBeacon device and the smartphone (b) Monitoring the dataof smartphone devices that is the on-board camera andaccelerometer so as to record the behavior of taking picturesand standing still to appreciate something on each interestingspot of the tourist
To record the number of taking pictures behaviorsthe client App monitors the on-board camera operationmessage of Android system namely ldquoandroid hardwareactionNEW PICTURErdquo once the tourist uses the phonecamera to take a picture To record the number of standingbehaviors the client App integrates 3-dimensional acceler-ations into an overall acceleration data first Then it uses aSliding Window Filtering method [55] to count the numberof standing behaviors The client App inserts the number ofthese two behaviors into the current travel behavior sequenceLast the client App uploads the behavior sequence and itscorresponding profile to the system server when it detects theexit of the scenic area
Let 119861 = 1198871 1198872 119887119892 be the set of iBeacon devices thatare installed in a specific scenic area In the system servera travel behavior sequence record is stored as ltsid tbsgtwhere sid is the identifier of the sequence and tbs is an onsitebehavior sequence And tbs consists of a sequence (ltstin1 b1stout1 p1 s1gt ltstin2 b2 stout2 p2 s2gt ltstink bk stoutkpk skgt) where the quintupleltstini bi stouti pi sigt representsa behavior data with respect to the interesting spot i bi is thecorresponding iBeacon device ID and 119887119894 isin 119861 stini and stoutistands for arriving and leaving timestamps respectively andstinilestoutilestini+1 for 1 le 119894 le 119896 minus 1 pi and si are the numberof taking pictures and standing still to appreciate somethingrespectively Further the visit duration of spot i is calculatedby stouti - stini the interval between spot i and spot i+1 iscalculated by 119904119905119894119899119894+1 minus 119904119905119900119906119905119894Example 1 As illustrated in Figure 1 there are one entranceone exit and seven interesting spots in the scenic area Thusthere are nine iBeacon devices as total needed to install inthe area After tourist 4 inputs his or her profile and timeconstraint the system returns a travel route by mining thehistorical travel behavior sequences acquired from the otherthree tourists The corresponding sequences are shown inTable 1 for example tourist 1 visited six interesting spots ABD FE andGThe symbolsZe andZt stand for the entranceand the exit respectively Taking the behavior data at spot Aas an instance tourist 1 arrived at spot A at the 6thmin andleft out at the 26th min took 5 pictures and stood still for 4times at spot A
33 Tourist-Behavior Mining The goal of the Tourist-Behavior mining stage is to generate various candidatetravel routes by mining the historical onsite travel behaviorsequences This stage consists of two steps the travel behav-ior sequence preprocessing step and the Tourist-Behaviorsequential travel routes generating step
331 Travel Behavior Sequence Preprocessing The prepro-cessing step is to transform travel behavior sequences intoTourist-Behavior (TB) pattern sequences and then storepattern sequences into route subset according to their cor-responding personal profile Before describing the details ofthe step the following definitions are given
Definition 2 ATourist-Behavior (TB) pattern 120582119894 is defined asa triple ltbi NPi Di gt where bi is the location identity of spoti NPi is the normalized popularity value about spot i Di isthe discrete visit duration at spot i Note that the pattern 120582119894 issaid to match the pattern 120582119895 if and only if bi = bj NPi = NPjand Di = Dj
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
6 Wireless Communications and Mobile Computing
Definition 3 Let Λ = 1205821 1205822 120582119909 be the set ofTB patterns and let 120575119894 be the discrete interspot traveltime in a travel behavior sequence A sequence 120572 =(1198811 1205751 1198812 1205752 120575119896minus1 119881119896 ) is a TB sequence if 119881119904 isin Λ for1 le 119904 le ℎ and 120575119904 = 119863119894119904119888119879(Δ119905) for 1 le 119904 le ℎ minus 1
First the preprocessing method cleans up the passing-bybehavior data and calculates the interspot travel time and thevisit duration in each travel behavior sequence Hence themethod needs to delete the behavior data if the visit durationis shorter than a time threshold Tv except for the entranceand exit behavior data Let 120575119894 be the discrete interspot traveltime for the tourist to travel from spot i to spot i+1 and letDibe the visit duration at spot I Δ119905 stands for 120575119894 or Di and Tdis the metric of the discrete time Consequently the discretetime integer of 120575119894 and Di can be derived from the followingequation
119863119894119904119888119879 (Δ119905) = lceilΔ119905119879119889 rceil (2)
Second the method calculates popularity values of eachinteresting spot in each travel behavior sequence As eachtravel behavior sequence is collected from an individualtourist two popularity values of the same spot in twosequences are probably different due to two different touristsrsquoonsite behaviors The prior knowledge of the method is thattourists will spend longer visit duration take more picturesor stand still more times to appreciate something at a spot ifthey are more interested in the spot The popularity value ofspot i in a specific sequence can be calculated by the followingequation
119875119900119901119894 = 1199081 times 119904119905119900119906119905119894 minus 119904119905119894119899119894sum119896119894=1 (119904119905119900119906119905119894 minus 119904119905119894119899119894) + 1199082 times119901119894sum119896119894=1 119901119894 + 1199083
times 119904119894sum119896119894=1 119904119894(3)
where 1199081 1199082 and 1199083 are weights used to calculate Popiand 1199081 + 1199082 + 1199083 = 1 for each travel behavior sequencethe total visit duration is derived from sum119896119894=1(119904119905119900119906119905119894 minus 119904119905119894119899119894)sum119896119894=1 119901119894 andsum119896119894=1 119904119894 denote the total number of times of takingpictures and standing still within the sequence respectivelyBesides to make the popularity values of spots in differentsequences comparable we normalize all popularity values ineach sequence In detail to calculate a normalized popularityvalue NPi of spot i all spots in each sequence are ranked as adescending list according to their respectivePopi To calculateNPi the list is divided into n segments where n denotes thepopularity normalization coefficient For example in (4) thenormalization coefficient is to be 4 all spots ranking in thetop 1n in a specific sequence are assigned with a normalizedpopularity value of Ln indicating that the querying tourist is
most likely to be interested in these spots
119873119875119894 =
119871119899 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119905119900119901 11198991198712 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 [119899 minus 2119899 119899 minus 1119899 )1198711 119894119891 119875119900119901119894119903119886119899119896119904 119894119899 119905ℎ119890 119897119900119908119890119904119905 1119899
(4)
After the preprocessing step all travel behavior sequencesare transformed into TB pattern sequences and stored in theTourist-Behavior sequence database (abbr TBD) accordingto their respective profiles Specifically our system dividestourist age into three groups below 20 years from 20 to 55years and over 55 years and classifies education into threelevels preundergraduate undergraduate and graduate Bymultiplying with two gender attributes there are 3 times 3 times 2 =18 TBDs in total in our system with the above three profileattributes
Example 4 Let us take the travel behavior sequences shownin Table 1 as an example to explain the travel behaviorsequence preprocessing method Suppose that Tv is set at 5minutes Td is set at 10 minutes and 1199081 1199082 and 1199083 are setas 04 03 and 03 respectively At first the behavior datalt99 E 101 0 0gt is deleted as a passing-by behavior data insid 01 because its visit duration is shorter than Tv Furtherthe interspot travel time and the visit duration are discretizedby (2) Next the popularity of each spot is computed forexample PopA with respect to tourist 1 is calculated as 04 times2082 + 03 times 514 + 03 times 418 = 0271 the visit durationDA is 20 minutes the total visit duration is 82 minutes thenumber of taking pictures and standing still is 14 and 18respectively The corresponding TB pattern sequences areshown in Table 2
332 Tourist-Behavior Sequential Travel Routes GeneratingAs the onsite travel behaviors are complex and contain noisybehavior data for example onemaking a phone call or takinga sit for a break during a visit we need a method to discoverpopular travel routes and to filter noise travel behaviorsTherefore we design the TB PrefixSpan algorithm to discoverall frequent TB patterns with the corresponding interspottravel time and to construct various Tourist-Behavior (TB)sequential travel routes from a TBD An improvement ofthe TB PrefixSpan algorithm compared to [54] is that dueto the fact that TB pattern sequences separately containdiscrete interspot travel time and spot visit durations theTB PrefixSpan algorithm can delete visit durations of non-frequent TB patterns yet preserve intervals to ensure theaccurate time arrangement of new TB sequential patternsBefore describing the TB PrefixSpan algorithm the followingdefinitions are given
Definition 5 Assume two TB pattern sequences 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is said to be contained in 120572 or aTB subsequence of 120572 that is 120573 sube 120572 if there exist sequence
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 7
Table 2 An example of Tourist-Behavior pattern sequence
Sid Tourist-Behavior sequence01 (ltZe L1 1gt1ltA L4 2gt1ltB L2 1gt1ltD L2 2gt1ltF L3 2gt2ltG L3 2gt1lt Zt L1 1gt)02 (ltZe L1 1gt1ltA L42gt1ltB L2 1gt1ltC L2 1gt2ltF L3 2gt1ltG L3 2gt1lt Zt L1 1gt)03 (ltZe L1 1gt1ltA L2 2gt3ltC L32gt3ltF L3 2gt1ltE L4 2gt1ltG L2 2gt1lt Zt L1 1gt)
Table 3 An example of TB sequences database
Sid TB pattern sequences01 (ltZeL11gt1ltAL43gt1ltCL32gt2ltEL32gt1ltFL22gt1ltGL42gt2ltHL43gt1ltLL21gt1ltXL23gt1ltZtL11gt)02 (ltZeL11gt1ltAL43gt1ltBL33gt1ltCL31gt2ltFL23gt1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)03 (ltZeL11gt1ltAL33gt1ltBL33gt2ltEL23gt1ltGL42gt2ltLL42gt1ltDL32gt1ltXL23gt1ltZtL11gt)04 (ltZeL11)1ltAL22gt2ltDL32gt1ltEL32gt2ltHL43gt2ltLL21gt2ltRL33gt1ltXL23)1ltZtL11gt)05 (ltZeL11gt1ltAL43gt2ltCL31gt2ltFL23gt1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
indices 1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 such that (1) 1198811205731 = 1198811205721198951and 1198811205732 = 1198811205721198952 119881120573ℎ = 119881120572119895ℎ (2) 1205751205731 = 1205751205721198951 and 1205751205732 =1205751205721198952 120575120573ℎ = 120575120572119895ℎDefinition 6 A TB pattern 120574 is called a frequent TBpattern if the number of sequences in a TBD whichcontains 120574 as the subsequence is greater than or equalto the user-specified minimum support called min supor min sup count That is 120574 is called a frequent TBpattern in a TBD if sup 119888119900119906119899119905119879119861119863(120574) ge |119879119861119863| times119898119894119899 119904119906119901 or119904119906119901 119888119900119906119899119905119879119861119863(120574) ge 119898119894119899 119904119906119901 119888119900119906119899119905 wheresup 119888119900119906119899119905119879119861119863(120574) = |120573119894 isin 119879119861119863 and 120574 sube 120573119894 1 le 119894 le |119879119861119863||Definition 7 Assume a TB pattern sequence 120572 = (1198811205721 12057512057211198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120572 is called a TB sequentialtravel route if all TB patterns in 120572 are frequent TB patternsfurther 120572 can be referred to as a k-length TB sequential travelroute
Definition 8 Given two TB sequential travel routes 120572 = (11988112057211205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a TB prefix of 120572 if and only if(1) 119881120573119894 = 119881120572119894 for 1 le 119894 le ℎ (2) 120575120573119894 = 120575120572119894 for 1 le 119894 le ℎ minus 1Definition 9 Given two TB sequential travel routes 120572 =(1198811205721 1205751205721 1198811205722 1205751205722 120575120572(119896minus1) 119881120572119896) and 120573 = (1198811205731 12057512057311198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119896) 120573 is a subsequence of 120572 Let1 le 1198951 lt 1198952 lt sdot sdot sdot lt 119895ℎ le 119896 be the indices of frequent TBpatterns contained in 120572 which match in 120573 A subsequence1205721015840 = (11988112057210158401 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119892minus1) 1198811205721015840119892) of 120572 where 119892 =ℎ + 119896 minus 119895ℎ is named a projection of 120572 with respect to 120573 if andonly if (1) 120573 is a TB prefix of 1205721015840 and (2) the last 119896 minus 119895ℎ TBpatterns of 1205721015840 are the same as the last 119896 minus 119895ℎ TB patterns of 120572Definition 10 Let 1205721015840 = (1198811205721015840 12057512057210158401 11988112057210158402 12057512057210158402 1205751205721015840(119898minus1) 1198811205721015840119898)be the projection of 120572 with respect to a TB prefix 120573 =(1198811205731 1205751205731 1198811205732 1205751205732 120575120573(ℎminus1) 119881120573ℎ)(ℎ le 119898) Then 120579 =(1198811205721015840(ℎ+1) 1205751205721015840(ℎ+1) 1198811205721015840(ℎ+2) 1205751205721015840(ℎ+2) 1205751205721015840(119898minus1) 1198811205721015840(119898)) is theTB postfix of 120572 with respect to prefix 120573
The pseudocode of the TB PrefixSpan algorithm is shownin Figure 3The 120572ndashprojection database consists of postfixes of
TB pattern sequences in a TBD with respect to the TB prefix120572 which is denoted as TBD|120572 As the original PrefixSpanalgorithm does not include the relationship among two TBpatterns and their interval a TB Table is designed to store thistype of relation where a row corresponds to a TB pattern anda column corresponds to a 120575 value For instance TB Table|120582119894stores the support count of subsequences with respect to thecurrent TB prefix 120572which has the last TB pattern 120582119894The tablecell TB Table|120582119894(120575119873 120582119896) records the number of subsequencesin TBD|120572 containing the TB pattern subsequence (120582119894 120575119873 120582119896)Note that 120575119873 is an accumulated time from spot i to spot k thatis 120575119873 = 120575119894 + 120575119894+1 + sdot sdot sdot + 120575119896minus1
Specifically the algorithm initially recognizes each fre-quent TB pattern to construct their corresponding 120572-projection databases For each TBD|120572 database the algorithmconstructs the corresponding TB Table to identify all fre-quent table cells Then for each frequent cell the element(120575119873 120582119895) is appended to the end of 120572 to construct a newTB prefix 1205721015840 and then the 1205721015840-projection database TBD|1205721015840 isbuilt Recursively constructing all of the frequent TB patternsequences in the TBD|1205721015840 discovers all TB sequential travelroutes in theTBDwhich are stored in theTB sequential travelroute database (abbr TBSTR)
Example 11 Let us take five TB pattern sequences in Table 3as an example to explain the TB sequential travel routemining process where the min sup count is set at 2 The TBPrefixSpan algorithm can be also deemed as a Tree Traversalalgorithm each node of the growing tress corresponds toa TB Table|120582119894 As shown in Figure 4 we use red solidarrows to illustrate steps of constructing a TB sequentialtravel route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt 4 ltXL32gt1 ltZtL11gt) In the beginning the algorithm constructsTB Table|ltgt which recognizes 16 1-length sequential routeslisted in Table 4 Suppose that the algorithm is mining thecurrent 3-length route (ltZeL11gt 1 ltAL43gt 2 ltFL23gt)Thus TB Table|lt11986511987123gt needs to be constructed subsequentlyAs shown in Table 5 the ltFL23gt-projection database hastwo TB postfix sid 02 and 05 The results of TB Table|lt11986511987123gtare shown in Table 6 There are 3 frequent TB patternslabeled in bold in the table that is ltHL22gt ltXL32gt
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
8 Wireless Communications and Mobile Computing
Figure 3 The pseudocode of the TB PrefixSpan algorithm
and ltZtL11gt The algorithm joins these 3 patterns behindthe current route respectively to construct 3 new 4-lengthroutes and then constructs 3 corresponding TB Tables ofthese patternsThe algorithm recursively traverses the tree todiscover all potential TB sequential travel routes
34 Travel Route Ranking and Recommending Asmentionedin Section 331 all TB pattern sequences are divided intosubsets according to the personal profile after the sequencepreprocess step To make the recommended routes matchthe querying touristrsquos personal interests and characteristicsbetter we design a route ranking method to search valuable
Table 4 An example of 1-length frequent TB pattern
TB pattern Sup count TB pattern Sup countltZeL11gt 5 ltHL22gt 2ltAL43gt 3 ltHL23gt 2ltBL33gt 2 ltLL21gt 2ltCL31gt 2 ltLL42gt 2ltDL32gt 2 ltXL23gt 3ltEL32gt 2 ltXL32gt 2ltFL23gt 2 ltRL33gt 2ltGL42gt 2 ltZtL11gt 5
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 9
lt gt
(ltAL43gt)hellip(ltZeL11gt)hellip (ltXL23gt)
TB_TABLE|
(1ltAL43gt)hellip(1ltBL33gt)
TB_TABLE|
hellip (2lt FL23 gt) hellip
TB_TABLE|
hellip
TB_TABLE|
hellip
TB_TABLE|
(4ltXL32gt)(1ltHL22gt)(5ltZtL11gt)
TB_TABLE|
(3ltXL32gt)hellip
TB_TABLE|
(1ltZtL11gt)
TB_TABLE|
Oslash
hellip
hellip
hellip
ltgt
ltZeL11gt
ltAL43gt
ltXL23gtltAL43gt
ltFL23gt
ltHL22gtltXL32gt
Figure 4 An example of TB sequential travel routes generation process
Table 5 An example of the ltFL23gt-projection database TBD|lt11986511987123gtSid TB Pattern ltFL23gt-projection database02 (1ltHL22gt1ltLL42gt1ltRL33gt1ltXL32gt1ltZtL11gt)05 (1ltHL22gt2ltRL31gt1ltXL32gt1ltZtL11gt)
Table 6 An example of TB Table|120582119894TB pattern Interspot120575119873
1 2 3 4 5ltHL22gt 2 0 0 0 0ltLL42gt 0 1 0 0 0ltRL33gt 0 0 1 0 0ltRL31gt 0 0 1 0 0ltXL32gt 0 0 0 2 0ltZtL11gt 0 0 0 0 2
and reasonable routes from a TBSTR matched by an inputpersonal profile
Thus the method first requests the querying tourist toinput a personal profile and a route constraint The routeconstraint includes the intended travel duration and specifiedtravel start and end location of the POI Next the methoduses the personal profile to retrieve candidate TB sequentialtravel routes in the corresponding TBSTR After retrievingcandidate TB sequential travel routes the server filters outtravel routes that do not meet the input route constraintIn detail the server reserves the routes of which start andend points match the user-specified entrance and exit Thisis for considering the situation of multiple entrances andexits existing in one scenic area Then it adds up the total
route duration time 119879119905120572 of the route 120572 by using the followingequation
119879119905119886 = (|120572|minus1sum119894=1
120575119894 + |120572|sum119894=1
119863119894) times 119879119889 (5)
where |120572| is the length of the route 120572 120575i and Di are theith discrete interval and spot visit duration of the route 120572respectively Td is the metric of the discrete time And theserver selects the candidate routes that meet the followingequation
(1 minus 120593) times 119868119879119863 le 119879119905119886 le 119868119879119863 (6)
where ITD stands for an intended travel duration of thequerying tourist 120593 is a filtering condition parameter between[0 1] used to set a filtering range for the candidate routes
At last the system server recommends the most valuableTop-k tangible travel routes for the querying tourist bycalculating route values of the remaining routes A route valueconsists of the total normalized popularity value and the ratioof the total visit duration to the total route duration Thecandidate travel routes set is denoted as 119862119877119894 | 119894 = 1 2 119872
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
10 Wireless Communications and Mobile Computing
An iBeacon device
A smart phone running a client App
(a)
The recognized spot
(b)
Camera broadcast message Acceleration
Timestamp
Major Minor
(c)
Figure 5 A test of onsite behavior acquisition
where M is the total number of candidates The route valueRV120572 of route 120572 is calculated by the following equation
119877119881120572 = 119908119901119897 times119872119872119873( |120572|sum119894=1
119873119875119894) + 119908119903V119889
times ( sum|120572|119894=1119863119894sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) (7)
where 119908119901119897 and 119908119903V119889 are the weights used to calculate 119877119881120572and119908119901119897 +119908119903V119889=1sum|120572|119894=1119873119875119894 is the total normalized popularityvalue of route 120572 (sum|120572|119894=1119863119894)(sum|120572|minus1119894=1 120575119894 + sum|120572|119894=1119863119894) is the ratio ofthe total visit duration to the total route duration of route 120572MMN(sdot) is min-max normalization function for normalizingthe rank value among 119862119877119894 | 119894 = 1 2 1198724 Experiment and Discussion
In this section we designed a validation experiment andseveral performance analysis experiments to test our systemThe iBeacon devices were based on CC2541 embedded pro-cessor developed by Texas Instruments Company The clientApp was developed with Android Studio 233 which runson Android smartphone system version 601 upwards Thesystem server runs on a workstation with Intel Xeon 35GHzand 16 GB RAM and related application was developed bypython version 2713 running on Ubuntu 1404 with Tomcat60
41 Validation Experiment The primary goals of our empiri-cal experiment are to (1) examine howwell the recommendedroutes match a touristrsquos actual interests (2) demonstratethe route value of the top recommended routes and (3)analyze the rationality of the top recommended routes Inthis section we introduce the experimental settings of the
validation experiment and present the results of a test ofonsite behavior data collection Last we present the resultsof the validation experiment to validate the ability of oursystem in recommending personalized tangible travel routesfor tourists in a given POI based on historical onsite travelbehavior
At first we deployed our system in a small experimen-tal exhibition hall with 20 exhibits where each exhibit ispreinstalled with an iBeacon device And we invited 20male and 20 female undergraduate students as volunteersto visit the experimental hall so as to collect their onsitetravel behavior data The layout of the experimental hall isshown as in Figure 6 in which topical-related posters areexhibited with a similar length In detail from B1 to B5 arescientific topical posters from B6 to B12 are sports-relatedand from B13 to B20 are daily life related contents Andwe set the minimum support count of TB PrefixSpan at 2and set discrete time metric Td at 1 minute the popularitynormalization coefficient is to be 5 the filtering conditionparameter 120593 is set at 02 the weights 119908119901119897 and 119908119903V119889 in (7) wereequally set at 05
Figure 5 illustrates a test of onsite behavior acquisitionFigure 5(a) shows the experimental environment in whichone volunteer carrying a smartphone is visiting an exhibitthat is labeled by an iBeacon device Figure 5(b) shows thesoftware interface of the client App which is selecting thenearest iBeacon device as the recognized spot that is Minor3 device to collect following onsite behavior data Figure 5(c)presents the original behavior sequence gathered on theMinor 3 spot
To examine how well the recommended routes match atouristrsquos actual interests we requested the volunteers to filla rating questionnaire regarding the exhibiting posters aftervisiting the exhibition hall so as to directly learn interests offemale and male volunteers Table 9 lists the most favorite10 posters for female and male volunteers respectivelywhich reflects that female volunteers prefer daily life topical
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 11
Ei EntranceExit
B1
B2
B3 B4
B5B7
B6
B9
B10 B11
B12
B13
B14 B17
B18 B19
Bi the i-th exhibit
B8 B15 B16 B20
E0
E1
Female Route
Male Route
Figure 6 The result of top 1 tangible travel route for two queryingtourists
Table 7 Two querying touristsrsquo route constraints and personalprofiles
Tourist Time constraint Personal profile1 45 min YoungFemaleUndergraduate2 30 min YoungMaleUndergraduate
posters while male volunteers prefer sports-related contentsNext we assumed two querying touristsrsquo route constraintsand personal profiles that are listed in Table 7 to requestroute recommendations from our system The top 3 valuabletangible travel routes recommended to two tourists are listedin Table 8 It can be easily observed from Table 8 that allcandidate routes comply with corresponding touristrsquos timeconstraints More importantly by comparing the posters ofTables 8 and 9 we find that about 80 of top 10 favoriteposters regarding the two corresponding tourists are includedin both top recommended routes For instance the first routerecommended to the female tourist suggests she spends arelatively longer time at B2 B15 and B16 posters which are thetop favorite posters to female listed in Table 9 while the firstroute recommended to the male tourist recommends B4 B6and B7 posters This observation proves that our system canlearn different personal preferences from real onsite travelbehaviors
To demonstrate the route value of the top recommendedroutes our route ranking method ranks top 3 valuable tangi-ble travel routes which are listed in Table 8 It can be easilyobserved in Table 8 that both top 1 routes recommended toyoung female and male tourists have the biggest route valueFor example compared to the rest two routes the top 1 routerecommended to young male tourist possesses the largesttotal normalized popularity value (ie L38) the longest totalvisit duration (ie 35minutes) and the largest number of visitspots (ie 8 spots)
Regarding the rationality of the top recommended routesas the recommended tangible routes are generated fromonsite travel behaviors of young female or male touriststhe visit arrangements of the recommended route (eg thevisit sequence the interspot travel time and the spot visit
3223 3472 3635 3757 3845
5 5
7 7 7
500 1000 1500 2000 2500Number of sequences
Average lenth
Longest length
1
2
3
4
5
6
7
8
Leng
th o
f tra
vel r
oute
Figure 7 The length of TB sequential travel routes under variousdata sizes
duration) can completely comply with the layout of theexhibition hall As a result the recommended routes possessrather visit rationality For instance the interspot travel timeof a tangible route recommended to young females is derivedfrom the average walking speed of young females and thevisit duration of each spot is calculated from the historicalonsite travel behaviors of young females for example theaverage reading speed of young females Figure 6 illustratesthe visit sequence of two top 1 tangible routes for twoqueryingtouristsThe results indicate that the recommended route canhelp two querying tourists to finish their respective time-limited visits comfortably
42 Algorithm Performance Analysis In this section westudy the performance of our TB PrefixSpan algorithmunder different parameters settings Obviously longer travelroutes which containmore interesting spots canmeet longertravel duration query needs Meanwhile larger number ofgenerated travel routes can provide more diverse personalrecommendations for tourists Thus the length and thequantity of generated TB sequential travel routes reflect theeffectiveness and quality of the recommendations in thiswork Furthermore to test the scalability of our algorithm weconstruct a synthetic data set consisting of 12000 randomlygenerated travel behavior sequences All of the experimentaltravel behavior sequences are randomly selected from thesynthetic data set Without any other notice the followingexperimental parameters settings are the same as the valida-tion experiment
421 Data Size of Travel Behavior Sequences To comprehendhow the number of travel behavior sequences (data size)affects the TB sequential travel route generation the datasize is changed from 500 to 2500 and the min sup is setat 4 Figure 7 demonstrates the average length and thelongest length of the generated routes under different datasizes As the data size is getting bigger the length of TBsequential travel routes is getting longer that is the quality ofroutes is getting better Therefore the more historical onsite
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
12 Wireless Communications and Mobile Computing
Table8To
p3cand
idatetangibletravelrou
tesg
enerated
fortwotourists
Tourist
Cand
idatetangibletravelroutes
Totalroute
duratio
nTotalvisit
duratio
nTotal
exhibits
Total119873119875119894
1ltE0
L1-gt1lt
B1L44gt1
ltB2L55gt0
ltB3L43gt1
ltB6L42gt2
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B17L
45gt1
ltE1L1-gt
43min
35min
8L3
8ltE0
L1-gt1lt
B2L55gt1
ltB3L43gt1
ltB12L4
2gt2
ltB14L54gt1lt
B15L56gt1lt
B17L
45gt1
ltB16L56gt1lt
E1L1-gt
40min
31min
7L3
4ltE0
L1-gt1lt
B1L44gt1
ltB4L44gt0
ltB5L32gt3
ltB14L54gt1lt
B15L56gt0
ltB16L56gt1lt
B20L32gt1lt
E1L1-gt
36min
28min
7L31
2ltE0
L1-gt1lt
B4L54gt0
ltB5L43gt1
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt0
ltB17L
42gt1
ltE1L1-gt
28min
25min
7L3
4ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB7L55gt0
ltB12L54gt0
ltB14L4
3gt1
ltB15L4
2gt2
ltE1L1-gt
26min
22min
6L3
0ltE0
L1-gt1lt
B4L54gt0
ltB6L54gt0
ltB9L56gt1
ltB8L43gt1
ltB17L
42gt1
ltE1L1-gt
23min
19min
5L2
5
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 13
Table 9 The top 10 favorite posters of two types of volunteers
Young Female Young MalePoster No Poster Topic Poster No Poster TopicB2 Scientific B4 ScientificB14 Daily life B6 SportB1 Scientific B7 SportB3 Scientific B12 SportB15 Daily life B9 SportB16 Daily life B5 ScientificB17 Daily life B8 SportB19 Daily life B14 Daily lifeB12 Sport B2 ScientificB11 Sport B17 Daily life
5144476 4196 406 398
11
87
6 6
1
3
5
7
9
11
13
002 004 006 008 010
Leng
ht o
f tra
vel r
oute
s
Minimum supportAverage length
Longest length
Figure 8 The length of TB sequential travel routes under differentminimum supports
travel behavior the system gets the higher quality of therecommendation can be generated
422 Minimum Support of TB PrefixSpan To discover howthe minimum support parameter min sup of TB PrefixSpanaffects the quality of the TB sequential travel route thedifferentmin sup varying from 002 to 01 are tested withthe synthetic data set
Figure 8 presents the average length and the longestlength of the generated routes under differentmin sup As theminimum support increases the length of generated routesdecreases If the min sup is set at 002 the average lengthof routes is 514 and the longest route contains 11 visit spotsHowever if the min sup is set at 01 the average length ofroutes declines to 398 and the longest route only contains 6visit spots
Figure 9 shows the execution time of the TB PrefixS-pan algorithm under different minimum supports As themin sup increases from 002 to 01 the execution timeof the TB sequential travel route generation process declinesfrom 95059s to 2005s Based on the observation fromFigures 8 and 9 themin sup is suggested as 004 or below toensure the quality of the routes As the algorithm is performed
95059
4154831421
23766 2005
002 004 006 008 010Minimum support
0
200
400
600
800
1000
The e
xecu
tion
time (
seco
nd)
Figure 9The execution time of the TB PrefixSpan algorithm underdifferent minimum supports
331388 405 434
5
7 78
123456789
5 10 15 20
Leng
ht o
f tra
vel r
oute
Average length
Longest length
Metric of the dicrete time 4>
Figure 10 The length of TB sequential travel routes with differentTd
in an offline stage on the system server side the running timeof the algorithmwill not affect the reaction speed of the onlinerecommendation process
423 Information Granularity of Tourist-Behavior PatternAccording toDefinition 2 eachTBpattern includes a locationidentity a normalized popularity value and visit durationThus the information granularity of a TB pattern which isaffected by the metric of the discrete time and the popularitynormalization coefficient consequently decides the qualityof the generated travel routes To observe the relationshipbetween the information granularity and the route qualitythat is the average length and longest length of the generatedTB sequential travel routes the following experiments areconducted
Regarding the metric of the discrete time Td a series ofvalues ranging from 5 min to 20 min are tested by 2000sequences randomly selected from the data setWith differentTd settings the average length and the longest length of thegenerated routes are shown in Figure 10 and the number ofgenerated routes is shown in Figure 11When themetric of Tdis increasing both the length and the number of generatedroutes are increasing as well The reason is that if Td is
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
14 Wireless Communications and Mobile Computing
140892
261035
404742453895
0
100000
200000
300000
400000
500000
5 10 15 20
Num
ber o
f tra
vel r
oute
s
Metric of the discrete time 4>
Figure 11 The number of TB sequential travel routes with differentTd
4221358 3212 3099
7 7
6
5
4 6 8 10Normalization coefficient
Average length
Longest length
1
2
3
4
5
6
7
8
Leng
ht o
f tra
vel r
oute
Figure 12 The length of TB sequential travel routes with differentnormalization coefficients
getting bigger more TB patterns will be recognized as asame frequent TB pattern and lead to generating longer andbigger count of routes When Td increases however the timeprecision of the candidate routes is declined For exampleassume that the visit duration at a spot is 21 min If Td is setas 5 min then the discrete time is integer 5 the time error is 4min at the route recommendation phase IfTd is set as 20minthen the discrete time is integer 2 the time error increases to19 min Further the accumulating time error of the candidatetravel route is unacceptable under an improper Td valueTherefore to balance the relation between the quality of TBsequential patterns and the time error of the travel route Tdis suggested at 10 min in this work
Regarding the popularity normalization coefficient wetest the coefficient ranging from 4 to 10 segments with 2000randomly selected sequences Figures 12 and 13 illustrate thequality of the generated travel routes with different normal-ization coefficients As shown in Figures 12 and 13 when thecoefficient increases the length and the number of generatedroutes both decreaseThis is because that the larger coefficientis set the more TB patterns can be generated in a travelbehavior sequence That is more normalized popularityvalues will decrease the support count of the correspondingTB pattern However more normalized popularity values
507453
14748286936
64321
4 6 8 10Normalizaiton coefficient
0
100000
200000
300000
400000
500000
600000
Num
ber o
f tra
vel r
oute
s
Figure 13 The number of generated travel routes with differentnormalization coefficients
can describe a touristrsquos preference more precisely whichcan enhance the personalization of the recommendationTherefore to make a proper balance between the quality andthe personalization of the generated travel routes setting thenormalization coefficient as 5 is appropriate for this workFinally the observations of Figures 11 and 13 demonstratethat the proposed algorithm is effective in generating diversetravel routes
5 Conclusions
The main goal of our work is to design a travel route recom-mendation system that recommends personalized tangibletravel routes for various tourists within a given POI Firstof all we designed a novel method based on smartphoneand IoT infrastructure to collect onsite travel behaviors oftourists in a specific POI automatically To learn touristsrsquopreferences to each interesting object or spot we developedan Android App to record multiple onsite travel behaviorson each spot including visit duration taking pictures andstanding Next we designed a travel behavior sequence pre-processing method and a Tourist-Behavior sequential routemining algorithm to generate potential frequent tangibletravel routes Furthermore the route ranking method usesthe querying touristrsquos personal profile and route constraintto recommend personalized tangible travel routes Finallyexperimental results demonstrate that the proposed system isefficient and effective in recommending tangible travel routesbased on collected onsite travel behavior data
Our future works include (1) deploying our system in areal-world POI (2) using more types of smartphone sensorsto gather more types of onsite travel behaviors to learntouristsrsquo preference precisely (3) harnessing real-time con-gestion information at each spot of a scenic area to generatemore reasonable travel routes and further improve touristsrsquotravel experience and (4) using the current location andhistorical travel sequence of the querying tourist to generatea real-time route recommendation when the tourist requestsroute recommendations at an arbitrary location within a POIfor improving the flexibility of our system
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Wireless Communications and Mobile Computing 15
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This work was partially supported by the National NaturalScience Foundation of China (nos U1501252 61572146and U1711263) the Natural Science Foundation of GuangxiProvince (nos 2016GXNSFDA380006 and AC16380122)the Guangxi Innovation-Driven Development Project (noAA17202024) and the Guangxi Universities Young andMiddle-aged Teacher Basic Ability Enhancement Project (no2018KY0203)
References
[1] R Baraglia C I Muntean F M Nardini and F SilvestrildquoLearNext learning to predict tourists movementsrdquo in Proceed-ings of the 22nd ACM International Conference on Informationand Knowledge Management pp 751ndash756 2013
[2] D Lian V W Zheng and X Xie ldquoCollaborative filtering meetsnext check-in location predictionrdquo in Proceedings of the 22ndInternational Conference onWorld Wide Web pp 231-232 2013
[3] Q Liu SWu LWang andT Tan ldquoPredicting the next locationa recurrent model with spatial and temporal contextsrdquo in Pro-ceedings of the 30th AAAI Conference on Artificial Intelligencepp 194ndash200 2016
[4] Y Su X LiW Tang J Xiang andYHe ldquoNext check-in locationprediction via footprints and friendship on location-basedsocial networksrdquo in Proceedings of the 19th IEEE InternationalConference on Mobile Data Management pp 251ndash256 2018
[5] D Massimo M Elahi and F Ricci ldquoLearning user preferencesby observing user-items interactions in an IoT augmentedspacerdquo in Proceedings of the 25th ACM International Conferenceon User Modeling Adaptation and Personalization pp 35ndash402017
[6] SHHashemi and J Kamps ldquoExploiting behavioral usermodelsfor point of interest recommendation in smart museumsrdquo NewReview of Hypermedia and Multimedia vol 24 no 3 pp 228ndash261 2018
[7] S H Hashemi and J Kamps ldquoWhere to go next exploitingbehavioral user models in smart environmentsrdquo in Proceedingsof the 25th ACM International Conference on User ModelingAdaptation and Personalization pp 50ndash58 2017
[8] X Li G Cong X-L Li T-A N Pham and S KrishnaswamyldquoRank-geoFM a ranking based geographical factorizationmethod for point of interest recommendationrdquo in Proceedings ofthe 38th International ACM SIGIR Conference on Research andDevelopment in Information Retrieval pp 433ndash442 2015
[9] J Wang Y Feng E Naghizade L Rashidi K H Lim andK Lee ldquoHappiness is a choice sentiment and activity-awarelocation recommendationrdquo in Proceedings of the Companion ofthe e Web Conference pp 1401ndash1405 Lyon France 2018
[10] L Yao Q Z Sheng Y Qin X Wang A Shemshadi and QHe ldquoContext-aware point-of-interest recommendation using
Tensor Factorization with social regularizationrdquo in Proceedingsof the 38th International ACM SIGIR Conference on Researchand Development in Information Retrieval pp 1007ndash1010 2015
[11] G Cai K Lee and I Lee ldquoItinerary recommender system withsemantic trajectory pattern mining from geo-tagged photosrdquoExpert Systems with Applications vol 94 pp 32ndash40 2018
[12] P Bolzoni S Helmer K Wellenzohn J Gamper and P Andrit-sos ldquoEfficient itinerary planning with category constraintsrdquoin Proceedings of the 22nd ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems pp203ndash212 2014
[13] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized tour recommendation based on user interests and points ofinterest visit durationsrdquo in Proceedings of the 24th InternationalJoint Conference on Artificial Intelligence pp 1778ndash1784 2015
[14] C Bin T Gu Y Sun L Chang W Sun and L Sun ldquoPer-sonalized POIs travel route recommendation system based ontourism big datardquo in Proceedings of the Pacific Rim InternationalConferences on Artificial Intelligence (PRICAI) pp 290ndash2992018
[15] C-Y Tsai and S-H Chung ldquoA personalized route recommen-dation service for theme parks using RFID information andtourist behaviorrdquo Decision Support Systems vol 52 no 2 pp514ndash527 2012
[16] W Luo H Tan L Chen and L M Ni ldquoFinding time period-based most frequent path in big trajectory datardquo in Proceedingsof the SIGMOD Conference pp 713ndash724 2013
[17] C Tsai J J Liou C Chen and C Hsiao ldquoGenerating touringpath suggestions using time-interval sequential pattern min-ingrdquo Expert Systems with Applications vol 39 no 3 pp 3593ndash3602 2012
[18] J Pei J Han B Mortazavi-Asl et al ldquoPrefixSpan min-ing sequential patterns efficiently by prefix-projected patterngrowthrdquo in Proceedings of the 17th International Conference onData Engineering pp 215ndash224 2001
[19] K H Lim J Chan S Karunasekera and C Leckie ldquoTourrecommendation and trip planning using location-based socialmedia a surveyrdquo Knowledge and Information Systems pp 1ndash292018
[20] C Yun and M Chen ldquoMining mobile sequential patterns in amobile commerce environmentrdquo IEEE Transactions on SystemsMan and Cybernetics vol 37 no 2 pp 278ndash295 2007
[21] Y Chen and C Shen ldquoPerformance analysis of smartphone-sensor behavior for human activity recognitionrdquo IEEE Accessvol 5 pp 3095ndash3110 2017
[22] iBeacon for Developers 2019 httpsdeveloperapplecomibeacon
[23] J Borras A Moreno and A Valls ldquoIntelligent tourism recom-mender systems a surveyrdquo Expert Systems with Applicationsvol 41 no 16 pp 7370ndash7389 2014
[24] T Tsiligirides ldquoHeuristic methods applied to orienteeringrdquoJournal of the Operational Research Society vol 35 no 9 pp797ndash809 1984
[25] D Gavalas C Konstantopoulos K Mastakas and G PantziouldquoA survey on algorithmic approaches for solving tourist tripdesign problemsrdquo Journal of Heuristics vol 20 no 3 pp 291ndash328 2014
[26] D Gavalas V Kasapakis C Konstantopoulos G Pantziou andN Vathis ldquoScenic route planning for touristsrdquo Personal andUbiquitous Computing vol 21 no 1 pp 137ndash155 2017
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
16 Wireless Communications and Mobile Computing
[27] C ZhangH Liang andKWang ldquoTrip recommendationmeetsreal-world constraints poi availability diversity and travelingtime uncertaintyrdquoACMTransactions on Information and SystemSecurity vol 35 no 1 pp 1ndash5 2016
[28] C Zhang H Liang K Wang and J Sun ldquoPersonalized triprecommendation with POI availability and uncertain travelingtimerdquo in Proceedings of the 24th ACM International Conferenceon Information and Knowledge Management pp 911ndash920 2015
[29] D Gavalas V Kasapakis C Konstantopoulos G E PantziouN Vathis and C D Zaroliagis ldquoThe eCOMPASS multimodaltourist tour plannerrdquo Expert Systems with Applications vol 42no 21 pp 7303ndash7316 2015
[30] T Liebig N Piatkowski C Bockermann and K Morik ldquoPre-dictive trip planning-smart routing in smart citiesrdquo in Proceed-ings of the 2014 Joint Workshops on International Conference onExtending Database Technology EDBT 2014 and InternationalConference on Database eory pp 331ndash338 2014
[31] C Chen D Zhang B Guo X Ma G Pan and Z WuldquoTripPlanner personalized trip planning leveraging heteroge-neous crowdsourced digital footprintsrdquo IEEE Transactions onIntelligent Transportation Systems vol 16 no 3 pp 1259ndash12732015
[32] T Liebig N Piatkowski C Bockermann and K MorikldquoDynamic route planning with real-time traffic predictionsrdquoInformation Systems vol 64 pp 258ndash265 2017
[33] X Wang C Leckie J Chan K H Lim and T VaithianathanldquoImproving personalized trip recommendation by avoidingcrowdsrdquo in Proceedings of the 25th ACM International Confer-ence on Information and Knowledge Management pp 25ndash342016
[34] T Aoike B Ho T Hara J Ota and Y Kurata ldquoUtilising crowdinformation of tourist spots in an interactive tour recommendersystemrdquo in Information and Communication Technologies inTourism pp 27ndash39 Springer 2019
[35] K H Lim J Chan S Karunasekera and C Leckie ldquoPersonal-ized itinerary recommendation with queuing time awarenessrdquoin Proceedings of the 40th International ACM SIGIR Conferenceon Research and Development in Information Retrieval pp 325ndash334 2017
[36] Y Zheng Y Chen X Xie and W-Y Ma ldquoGeoLife20 alocation-based social networking servicerdquo Mobile Data Man-agement pp 357-358 2009
[37] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledge Discovery and Data Mining pp 1082ndash1090 ACM2011
[38] H Gao J Tang X Hu and H Liu ldquoExploring temporaleffects for location recommendation on location-based socialnetworksrdquo in Proceedings of the 7th ACMConference on Recom-mender Systems pp 93ndash100 2013
[39] BThomee D A Shamma G Friedland et al ldquoYFCC100M thenew data inmultimedia researchrdquoCommunications of the ACMvol 59 no 2 pp 64ndash73 2016
[40] K H Lim J Chan C Leckie and S Karunasekera ldquoPersonal-ized trip recommendation for tourists based on user interestspoints of interest visit durations and visit recencyrdquo Knowledgeand Information Systems vol 54 no 2 pp 375ndash406 2018
[41] A Majid L Chen H T Mirza I Hussain and G ChenldquoA system for mining interesting tourist locations and travelsequences from public geo-tagged photosrdquo Data amp KnowledgeEngineering vol 95 pp 66ndash86 2015
[42] T Guo B Guo J Zhang Z Yu and X Zhou ldquoCrowdtravelleveraging heterogeneous crowdsourced data for scenic spotprofiling and recommendationrdquo in Proceedings of the PacificRim Conference on Multimedia pp 617ndash628 2016
[43] S Jiang X Qian T Mei and Y Fu ldquoPersonalized travelsequence recommendation on multi-source big social mediardquoIEEE Transactions on Big Data vol 2 no 1 pp 43ndash56 2016
[44] G Hu J Shao F Shen Z Huang and H Tao Shen ldquoUni-fying multi-source social media data for personalized travelroute planningrdquo in Proceedings of the 40th International ACMSIGIR Conference on Research and Development in InformationRetrieval pp 893ndash896 2017
[45] K Ashton ldquoThat internet of things thingrdquoRFID Journal vol 22no 7 pp 97ndash114 2009
[46] A Zanella N Bui A P Castellani L Vangelista and M ZorzildquoInternet of things for smart citiesrdquo IEEE Internet of ingsJournal vol 1 no 1 pp 22ndash32 2014
[47] S H Ahmed and S Rani ldquoA hybrid approach smart street usecase and future aspects for internet of things in smart citiesrdquoFuture Generation Computer Systems vol 79 pp 941ndash951 2018
[48] C-Y Tsai M-H Li and R J Kuo ldquoA shopping behaviorprediction system considering moving patterns and productcharacteristicsrdquo Computers amp Industrial Engineering vol 106pp 192ndash204 2017
[49] H Tang S S Liao and S X Sun ldquoA prediction frameworkbased on contextual data to support mobile personalized mar-ketingrdquo Decision Support Systems vol 56 pp 234ndash246 2013
[50] D Cavada M Elahi D Massimo et al ldquoTangible tourism withthe internet of thingsrdquo in Information andCommunication Tech-nologies in Tourism pp 349ndash361 Springer Cham Switzerland2018
[51] A Kuusik S Roche and F Weis ldquoSMARTMUSEUM culturalcontent recommendation system for mobile usersrdquo in Proceed-ings of the 2009 Fourth International Conference on ComputerSciences and Convergence Information Technology pp 477ndash482IEEE Seoul South Korea 2009
[52] S Alletto R Cucchiara G Del Fiore et al ldquoAn indoor location-aware system for an IoT-based smart museumrdquo IEEE Internet ofings Journal vol 3 no 2 pp 244ndash253 2016
[53] H Guo L Chen G Chen and M Lv ldquoSmartphone-basedactivity recognition independent of device orientation andplacementrdquo International Journal of Communication Systemsvol 29 no 16 pp 2403ndash2415 2016
[54] C Tsai and B Lai ldquoA location-item-time sequential patternmining algorithm for route recommendationrdquo Knowledge-Based Systems vol 73 pp 97ndash110 2015
[55] C-H Lee C-R Lin andM-S Chen ldquoSliding-windowfilteringan efficient algorithm for incremental miningrdquo in Proceedingsof the 2001 ACM CIKM 10th International Conference onInformation and Knowledge Management pp 263ndash270 2001
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
International Journal of
AerospaceEngineeringHindawiwwwhindawicom Volume 2018
RoboticsJournal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Active and Passive Electronic Components
VLSI Design
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Shock and Vibration
Hindawiwwwhindawicom Volume 2018
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawiwwwhindawicom
Volume 2018
Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom
The Scientific World Journal
Volume 2018
Control Scienceand Engineering
Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom
Journal ofEngineeringVolume 2018
SensorsJournal of
Hindawiwwwhindawicom Volume 2018
International Journal of
RotatingMachinery
Hindawiwwwhindawicom Volume 2018
Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom