location recommender system

Download Location recommender system

Post on 15-Apr-2017

195 views

Category:

Data & Analytics

1 download

Embed Size (px)

TRANSCRIPT

Graduate thesis LOCATIONS RECOMMEND SYSTEM BASED ON USERS FEATURES USING HIDDEN MARKOV MODEL

Sinh vin: Bi nguyn tt thnh

20102149 - IS1 Vit nht K55

Gio vin hng dn: TS. V tuyt Trnh

Gi a im da trn phn tch thi quen tng tc ca ngi dng

Ngi dng a im Tng tc

Check-in

Cinema 2 -- 8:30

Coffee 3 -- 10:15

Museum 1 -- 11:00

Shopping 1 -- 11:45

Locations

User-Location Graph

Users

Trajectories

View

Want to go

Like

Check-in

Comment

Ni v cc h gi thng thy hin nay (foody, lozi, a im vit.)

nh ngha cc i tng c im i tng h tr g cho mc tiu n

2

Mc tiu bi ton

Gi a im cho ngi dng:

C nhn ha gi da trn

c im ca ngi dng

Hnh vi ngi dng

S thch v thi quen ca ngi dng

V tr ca ngi dng

Cht lng/ tin cy ca a im

Ni dung trnh by

t vn

Mc ch

Cch tip cn

Xy dng h thng

Thc nghim

Kt lun

M hnh ha i tng

Ngi dngSexBirthdayRelationJobAddressDistrictProvinceTng tcMc quan tmXem (view)1Mun n (wanttogo)2Thch(like)3 n(check in)4Nhn xt(comment)5a im addressdistrictprovincecategoriesopeningTimeclosingTimeminPricemaxPricelatitudelongitudetotalViewtotalWantToGototalLiketotalCheckIntotalComment

Vector ngi dng

Vector a im

AttributedistrictprovinceagesexrelationWeight0.250.2250.20.1750.15DimensiondistrictprovincelowerPriceupperPriceopeningTimeclosingTimeWeight0.2250.1750.1750.1750.1250.125DimensiontotalViewtotalWanttogototalLiketotalCheckintotalCommentWeight0.070.130.20.270.33

Vector thuc tnh a im th hin cc c im ca a im

Vector tng tc a im th hin cht lng a im

Biu din i tng bng vector

Pht biu bi ton

Input:

user profile

Interactions

Locations - profile

Processes

Output:

Cc a im c gi cho ngi dng

Interactions

Pht biu bi ton

Input:

user profile

Interactions

Locations - profile

Processes

Output:

Cc a im c gi cho ngi dng

Interactions

Ni dung trnh by

t vn

Mc ch

Cch tip cn

Xy dng h thng

Thc nghim

Kt lun

M hnh gi

Pha 1 Tng qut ha

Pha 2 C nhn ha

Xc nh tp ngi dng tng t

AttributeActual ValueStandardized ValueSEXnam100n0AGE0 ~ 15016 ~ 233523 ~ 3560> 35100..

DISTANCE

Khong cch hai vector c im

Chun ha d liu

a cc thuc tnh v cng n v v min gi tr

ACTIVITY

+ Mc hot ng tng tc.

+ Tnh da trn lng tng tc ca ngi dng.

Input:

Ngi dng U ( profile )

Tp ngi dng trong h thng

Output:

Us : Danh sch ngi dng xp theo tng t vi U gim dn

VD: Ba ngi dng u v ( u1, u2 )

=> xc nh u1 hoc u2 ging u

DISTANCE(u, u1) LA

tng t : (a im a im) :

DISTANCE :

Khong cch gia hai vector c im

QUALITY :

+ Chnh lch cht lng hai a im

+ Tnh da trn lng tng tc t ngi dng ti a im

SIMILARITY = DISTANCE * QUALITY_SUB

Pha 2

Tin x l d liu trc khi x l: chnh cc gi tr thuc tnh sai gi tr hoc khng c gi tr

To c s d liu Neo4j: cc node (ngi dng v a im) d dng biu din

To cc quan h gia ngi dng v ia im l cc tng tc

To cc quan h gia ngi dng, ngi dng l khong cch gia cc vector thuc tnh

Biu din d liu trng Neo4j

Khi to c s d liu

Ni dung trnh by

t vn

Mc ch

Cch tip cn

Xy dng h thng

Thc nghim

Kt lun

Thc hin

D liu kim th:

T trang foody.vn

D liu trong phm vi H Ni

User: 11550

Location: 14993

Vi mi a im gi L:

Chi ph gi : 10 im

Vi mi a im gi L: 6 trng hp nh gi

L l loi U thng tng tc nht: 10 im

L l loi U thng tng tc th hai : 7 im

L l loi U tng tc : 5 im

L l loi U thng tng tc nht nhng l a im U tng tc: 3 im

L l loi U tng tc nhng l a im U tng tc : 2 im

L l loi U cha tng tng tc: 0 im

Kt qu

Kt qu chp nhn c: 10, 7, 5

Kt qu t: 3, 2

Kt qu tht bi: 0

4 gi d liu: 100, 250, 500, 750 ngi dng a im t d lch s tng tc + 40% ngu nhin

Chn mt ngi dng bt k, thc hin gi ly 10 a im gi u tin xc nh chnh xc

Gi(USER)S a im / s ln thc hinKt qu 10 imKt qu 7 imKt qu 5 imKt qu 3 imKt qu 2 imKt qu 0 imTng imim chi ph100100 / 1038211221265951000250500 / 5018713290947832795000500700 / 702251491681211135419170007501000 / 1002912192554437154592410000Chp nhnTTht biGi 10071%3%26%Gi 25081.8%2.6%15.6%Gi 50077.43%3.29%19.28%Gi 75076.5%8.1%15.4%

22

Sn phm

(Windows phone 8.1 )

Around You : Gi a im ph hp vi ngi dng

Best Locations : a ra cc a im da vo xp hng cht lng a im

Find Location : Tm a im theo tn v xp hng theo cht lng

Ni dung trnh by

t vn

Mc ch

Cch tip cn

Xy dng h thng

Thc nghim

Kt lun

Kt qu t c

Hiu l thuyt v h gi : lc tng tc, lc ni dung v h thng lai

Hiu l thuyt v m hnh Markov n

Hiu l thuyt c bn v d liu th v cch s dng c s d liu Neo4j

Xy dng c h gi lai kt hp lc tng tc v lc ni dung

ng dng m hnh Markov n trong x l, c th l x l d liu thi gian

Xy dng thnh cng h gi a im vi d liu test t kt qu trung bnh: 50%

Lp trnh c bn trn nn tng di ng windows phone 8.1

Cc yu cu c bn ra c p ng trong h thng.

Ci thin, pht trin bi ton

Tng tnh chnh xc cho h thng bng cc gii php

Phn tch ngn ng t nhin cho cc nhn xt ca ngi dng

iu chnh cc tham s chun ha ca h thng

t thuc tnh thi gian cho tt c cc tng tc ca ngi dng

Tng hiu nng cho h thng bng cc gii php

Ci thin khi to mi quan h {user-user} khi xy dng c s d liu bng vic khoanh vng ngi dng tng t

p dng x l song song khi tnh ton trong pha tng qut

S dng c s d liu m tng tc qu trnh x l (Redis database)

CM N THY C V CC BN

CH LNG NGHE!

Cht lng ca mt a im

interaction_point

*100%

if (distance1 < distance2) {

if (distance1*6 > distance2*5) {

if (activity1*3 < activity2*2) {

result = 1;

} else return = -1;

} else return = -1;

}

if (distance1 > distance2) {

if (distance1*5 < distance2*6) {

if (activity1*2 > activity2*3) {

result = -1;

} else return = 1;

} else return = 1;

}

if (distance1 == distance2) {

if (activity1 < activity2) return = 1;

if (activity1 > activity2) return = -1;

if (activity1 == activity2) return = 0;

}

3 a im (u1, u2, u)

chn u1 hoc u2 tng t u hn

So snh 2 cp (u1, u) (u2, u)

return 1: u2 tng t u hn u1

return -1: u1 tng t u hn u2

return 0: u1, u2 cng tng t vi u

So snh tng t gia {ngi dng ngi dng}

Cinema 2

Museum 1

Coffee 3

Json ElementsStandardizing DataCreating Neo4J NodesCreating Neo4j Relationships {User-Location}Creating Neo4J Relationships {User-User}Read FileNeo4J Database

Json ElementsStandardizing DataCreating Neo4J NodesCreating Neo4j Relationships {User-Location}Creating Neo4J Relationships {User-User}Read FileNeo4J Database