social serendipity: mobilizing social...

28 PERVASIVEcomputing Published by the IEEE CS and IEEE ComSoc ■ 1536-1268/05/$20.00 © 2005 IEEE

T H E S M A R T P H O N E

Social Serendipity:Mobilizing SocialSoftware

Mobile phones have been adoptedfaster than any technology inhuman history and are nowavailable to the majority of peo-ple on Earth who earn more

than US$5 a day. More than 600 million phoneswere sold in 2004, many timesmore than the number of per-sonal computers sold that year.1

This new infrastructure ofphones is ripe for novel applica-

tions, especially given continual increases in theirprocessing power.

Many mobile devices also incorporate low-power wireless connectivity protocols, such asBluetooth, that can be used to identify an indi-vidual to other people nearby. We have developedan architecture that leverages this functionality inmobile phones—originally designed for commu-nication at a distance—to connect people acrossthe room. Serendipity is an application of thearchitecture. It combines the existing communi-cations infrastructure with online introductionsystems’ functionality to facilitate interactionsbetween physically proximate people through acentralized server.

Bluetooth proximity detectionAfter years of hype, Bluetooth is finally seeing

mass-market adoption in mobile electronics.According to the official Bluetooth Web site(www.bluetooth.com/news/releases.asp), more

than three million Bluetooth devices are currentlysold each week. The protocol’s primary use is toconnect items such as a wireless headset or laptopto a phone. As a by-product, Bluetooth devicescan also discover other devices nearby. This “acci-dental” functionality enables mobile communi-cation devices to act as a gateway to online intro-duction systems, such as Friendster, Monster, orMatch.com. The difference is that mobile devicessituate the introduction in an immediate socialcontext, rather than asynchronously in front of adesktop computer.

BlueAwareWe designed a Mobile Information Device Pro-

file (MIDP) 2.0 application, called BlueAware,that can run passively in the background on manyof today’s Bluetooth phones. This social-scanningapplication capitalizes on a unique Bluetoothidentifier (BTID) number that mobile phoneswith Bluetooth personal-area-network capabili-ties transmit when queried.

In a proximity log, BlueAware records and time-stamps the BTIDs it encounters. If it detects adevice that it has not recently recorded in the prox-imity log, it automatically sends the discoveredBTID to the Serendipity server.

Privacy-driven features. When a user turns on aBlueAware-equipped phone, the applicationbegins running in the background automatically,alerting the user to its presence with a dialog box

A new mobile-phone–based system uses Bluetooth hardware addressesand a database of user profiles to cue informal, face-to-face interactionsbetween nearby users who don’t know each other, but probably should.

Nathan Eagle and Alex PentlandMIT Media Laboratory

APRIL–JUNE 2005 PERVASIVEcomputing 29

at start-up. Additionally, the interfacedesign lets users read and delete the spe-cific data being collected, as Figure 1ashows, or stop the logging completely.

Refresh rate versus battery life. Continuallyscanning and logging BTIDs can expendthe battery of an older mobile phone—say, a two-year-old Nokia 3650—inabout 18 hours. While continuous scansprovide a rich depiction of a user’sdynamic environment, most individualsare used to having phones with standbytimes exceeding 48 hours. We thereforemodified BlueAware to scan the environ-ment only once every five minutes, pro-viding at least 36 hours of standby time.

BlueDarFigure 1b shows a variation on

BlueAware called BlueDar. We devel-oped it to be placed in a social settingand scan continuously for visible devices.BlueDar transmits detected BTIDs to theSerendipity server over an 802.11b wire-less network. It uses a specially designedBluetooth beacon that incorporates aclass 1 Bluetooth chipset controlled byan XPort Web server.2 We integrated thebeacon with an 802.11b wireless bridgeand packaged the components in anunobtrusive box.

We wrote an application that telnetscontinuously into multiple BlueDar sys-tems, scans repeatedly for Bluetoothdevices, and transmits the discoveredBTIDs to our server. Because the Blue-tooth chipset is a class 1 device, it candetect any visible Bluetooth device withina working range of about 30 meters.

Serendipity: Situatedintroductions

Today’s social software isn’t verysocial. From standard customer-rela-

tionship-management systems to Web-based introduction systems, these ser-vices require users to be in front of acomputer to make new acquaintances.Serendipity, on the other hand, embedssuch applications directly into everydaysocial settings: on the bus, around thewater cooler, in a bar, at a conference.

We use desktops, laptops, handheldcomputers, and mobile phones contin-ually in our work and social lives. Theseinnovations were primarily designed toempower the individual. However, overthe past decade, many instantiations ofsocial proximity sensing have appearedfor pocket-sized devices (see the side-bar, “Related Work in Mobile SocialSoftware”).

Mobile profile matchingSerendipity’s central server contains

user profiles along with matchmakingpreferences. The profiles are similar tothose stored in other social software pro-grams such as LinkedIn and Match.com.However, Serendipity users also provideweights that determine each piece ofinformation’s importance when calcu-lating a similarity score. The system cal-culates a similarity score by extractingthe commonalities between two proxi-mate users’ profiles and (if available)behavioral data, and summing themaccording to user-defined weights. If thescore is above the threshold set by bothusers, the server alerts them that some-one nearby might interest them.

Users can set thresholds from theirphones, along with the weightingschemes to define the similarity metrics.

The user-defined thresholds and weightscorrespond to profile types such as busi-ness networking, bar hopping, and busymodes. A patent on this behavioral/loca-tion-based matching and cueing systemwas filed in early 2004, and is now beinglicensed from MIT by SenseSix, a com-pany formed to commercialize this tech-nology.3

ImplementationSerendipity receives the BTID and

threshold variables from the phones andqueries a MySQL database for the userprofiles associated with the discoveredBTID addresses. If a profile exists, thesystem calls another script to calculate asimilarity score between the two proxi-mate users. When this score is aboveboth users’ thresholds, the script sendsan alert to their phones with the otheruser’s picture, their commonalities, a listof talking points, and additional contactinformation (at each user’s discretion).

A feedback mechanism lets users ratethe introduction message with a numbervalue from 1 to 10. Currently, only sys-tem designers are using this information,but it lays the foundation for a futurepersonalized matchmaking architecturebased on collaborative filtering.

Relationship inferenceFor approximately 100 users who

have opted in, we are also collecting avariety of additional data, including celltower IDs and communication logs,using the Context application (www.cs.helsinki.fi/group/context). Figure 2depicts the patterns in cell towers and

Figure 1. Methods for detecting Bluetoothdevices: (a) BlueAware running in theforeground on a Nokia Series 60 phone,and (b) BlueDar device using a Bluetooth beacon coupled with an 802.11b WiFibridge.

(a) (b)

30 PERVASIVEcomputing www.computer.org/pervasive


Bluetooth devices over the course of aday. It shows clearly that the user ishome from 1 a.m. to 10 a.m., then goesto work. We can see the majority ofdevices detected during regular officehours, with one exception—a device theuser comes into contact with outside theworkplace in the evening.

We use information about proximityto other mobile devices to make inferencesabout a user’s social network. Looking atthe proximity context between two peo-ple can reveal much information aboutthe nature of the relationship. For exam-

ple, proximity at 3 p.m. by the coffeemachines confers a much different mean-ing from proximity at 11 p.m. at a localbar. We have trained a Gaussian mixturemodel to detect proximity patternsbetween users and then correlate thesepatterns with relationship types.

The labels for this model came from asurvey that all experimental subjectstook at the end of two months’ data col-lection. The survey asked whom theyspent time with both in and out of theoffice and whom they considered to be intheir circle of friends. We compared these

labels with estimated location (using celltower distribution and static Bluetoothdevice distribution), proximity (mea-sured from Bluetooth logs), and time ofday; we discovered it is possible to iden-tify office acquaintances, outside friends,and people within a circle of friends withgood accuracy.4

Figure 3 illustrates some of the infor-mation that permits inference of friend-ship. It shows that our sensing techniqueis picking up the common-sense phe-nomenon that office acquaintances seeeach other frequently in the office but

M any applications of social proximity-sensing software have

appeared over the past decade. The following projects are

by no means a comprehensive review, but rather a sample of the

diversity in this burgeoning field.

LovegetyIntroduced in Japan in early 1998, Lovegety was the first com-

mercial attempt to move introduction systems away from the

desktop and into reality.1 Users input their responses to a couple of

questions into the Lovegety device, which then alerts them when

it senses a mutual match nearby.

Gaydar, a similar product specifically targeted for the gay com-

munity, appeared in the US soon after Lovegety’s launch.

Cell tower/SMS Friend FindersSeveral wireless service providers now offer location-based ser-

vices to mobile phone subscribers using cell tower IDs. For exam-

ple, Dodgeball.com users can expose their location to friends by

explicitly naming it through their short message service.

Experience Ubicomp ProjectCombining inexpensive radio frequency identification (RFID) tags

with traditional conference badges, the Experience Ubicomp Project

linked profiles describing many conference presenters with their

actual locations. When users would approach a tag reader, a nearby

screen would display “talking points” relevant to their interests.2

Social NetAnother project, Social Net, also uses RF-based devices—specifi-

cally, the Cybiko wireless communications computer (www.cybiko.

com)—to learn proximity patterns between people. When coupled

with explicit information about a social network, the device can

inform a mutual friend of two proximate people that an introduc-

tion might be appropriate.3

HummingbirdA custom-developed mobile RF device, the Hummingbird,4

alerts users when they’re near another user. The Hummingbird

supports collaboration and was designed to augment traditional

office communication media such as instant messaging and email.

JabberwockyA mobile phone application, Jabberwocky, performs repeated

Bluetooth scans to develop a sense of an urban landscape. By con-

tinually logging nearby phones, the system fosters a sense of the

“familiar strangers” that create an urban community.5 This func-

tionality distinguishes it from Serendipity, which focuses on intro-

ducing people directly.

REFERENCES

1. Y. Iwatani, “Love: Japanese Style,” Wired News, 11 Jun 1998; www.wired.com/news/culture/0,1284,12899,00.html.

2. J.F. McCarthy et al., “Proactive Displays and the Experience UbiCompProject,” ACM Siggroup Bull., Dec. 2002, pp. 38–41.

3. M. Terry et al., “Social Net: Using Patterns of Physical Proximity overTime to Infer Shared Interests,” Proc. Human Factors in Computing Sys-tems (CHI 2002), ACM Press, 2002, pp. 816–817.

4. L.E. Holmquist, J. Falk, and J. Wigström, “Supporting Group Collabora-tion with Interpersonal Awareness Devices,” J. Personal Technologies,vol. 3, nos. 1-2, 1999, pp. 105–124.

5. E. Paulos and E. Goodman, “The Familiar Stranger: Anxiety, Comfort,and Play in Public Places,” Proc. Human Factors in Computing Systems(CHI 2004), ACM Press, 2004, pp. 223–230.

Related Work in Mobile Social Software

rarely outside the office. Conversely,friends often see each other outside theoffice, even if they are also coworkers.

Other introduction servicesMost interactions instigated by

Serendipity are based on similarity scoresand information sent to two users. How-ever, proximity Web pages let users makepart of their profiles public and view thepublic information of other nearby users,regardless of similarity scores. With thisfeature, users need not disclose infor-mation about themselves to receiveinformation about others.

We have also enabled a feature thatsends only an anonymous text messagealerting users that a person who sharessimilar interests is nearby; both usersmust respond “yes” to actuate the dis-semination of any personal information.This feature preserves user privacy andminimizes disruptions.

Other methods exist for mediatingintroductions. For example, Social Netrelies on a mutual friend to make theintroduction (see the sidebar, “RelatedWork in Mobile Social Software”).Serendipity could incorporate such amethod, alerting a mutual friend—ratherthan the two individuals—when thealgorithm finds a match.

User studiesWe have tested and iterated the

Serendipity system for almost a year.

Initial deploymentIn early May 2004, Serendipity had its

first test deployment at a conference forsenior corporate executives and profes-sors. We created personal profiles for 40conference participants who picked uptheir assigned phones upon arrival in themorning. The system supported morethan 100 introductions over the courseof the day, primarily during the interses-sion coffee breaks. We used what welearned from this first deployment torefine the system in subsequent versions.

The conference setting necessitated

several modifications from our originaldesign. Because all the subjects wereproximate to each other during the talks,we had to develop a method for pre-venting introductions while the talkswere in session. Simply hard-coding theconference break schedule into phoneswas not advisable: first, the duration ofthe talks was uncertain; second, such anapproach would also prevent introduc-tions between people who were outsideduring a particular talk.

Instead, we used several of our researchgroup’s personal Bluetooth devices to pre-


Sun Mon WedTue SatFriThur(a)

Sun Mon WedTue SatFriThur(b)

0400 0800 1200 1600 2000

0400 0800 1200 1600

2400

24002000Figure 3. Inferring friendship: Contactdata from one subject for (a) a friend and(b) an office acquaintance. In each case,the two graphs show the proximityfrequency over a single day (top) and a single week (bottom).

(b)

BTID

0400

2468

101214

0800Time of day

Daily distribution of cell towers

1200 1600 2000

0400 0800 1200 1600 2000Time of day

Cell

tow

er ID

(a)

Cell

tow

er ID

Num

ber o

f tim

es a

dev

ice

was

dis

cove

red

Perc

enta

ge o

f tim

ein

a g

iven

ID

2468

101214

Time of day

Daily distribution of Bluetooth devices

0.90.80.70.60.50.40.30.20.1

108642

Figure 2. A sample user’s dailydistribution of (a) observed cell towertransitions and (b) Bluetooth devicedetections. The midday “hot spot” occurswhen the user is in the office.

vent intrusions. We had volunteers posi-tion themselves throughout the audito-rium, each carrying a visible Bluetoothdevice whose name was changed toBlock. Any of the 40 phones inside theauditorium during the talks could detectat least one of these Block devices. Whenit did so, it paused the application so thatthe Serendipity system neither recordedinformation about proximate devices norsent information to our server.

This succeeded in preventing intro-ductions during talks—when we knewthey were inappropriate. However, wehad not accounted for the density ofpeople mingling during breaks. Severalusers complained of receiving multipleintroductions within only a few min-utes of each other. This led to social dis-ruptions as one conversation just get-ting underway was interrupted by theinitiation of another. One user solvedthe problem by simply turning hisphone off while in conversation andthen turning it back on when he wasready to meet someone else. We subse-quently formalized this feature as Hid-den Mode. We also imposed a maxi-mum on receiving introductions to oneevery 10 minutes.

Among surprising results was theappreciation expressed by many userswho worked for large corporations atbeing introduced to other coworkers inthe same company. A couple of the par-

ticipants never understood the applica-tion’s introduction component; they didn’t know what the picture messagesabout people nearby were meant toaccomplish. However, aside from thecomplaints about disruptive multipleintroductions, the initial user feedbackwas primarily positive.

Few of the initial subjects voiced pri-vacy concerns, although many empha-sized the importance of being able to eas-ily turn the application on and off at theuser’s discretion.

Campus deploymentCurrently Serendipity is running on the

phones of 100 users at MIT. Of these, 70are either students or faculty in the sametechnical lab, while the remaining 30 areincoming students at the business schooladjacent to the laboratory. We are alsoreceiving information from the devicesregarding the other users that each userencounters throughout the day. The userprofiles from the technical lab are cur-rently bootstrapped from informationavailable within a public project direc-tory. Users can also input personal infor-mation and change any aspect of theirprofile. Figure 4 shows a portion of a userprofile.

Early user reactions have been over-whelmingly positive. Engineers and busi-ness school students have expressed thegreatest enthusiasm for introductions

based on their interest in the commer-cial potential of engineering researchprojects. The response to introductionsbetween members of the technical labhas also been positive. On average, thelab members are acquainted with onlyfive to seven other subjects in the study.Five percent of the users who receivedphones elected not to participate; the primary reason related to time—theydidn’t want to be interrupted—but twousers had privacy concerns as well.

BlueDar deploymentWe have networked BlueDar scanning

units in several social settings on campusincluding the student lounge, near the cof-fee machines, and at a local bar. Aboveeach device, we put a flyer explainingBlueDar’s functionality and the type ofdata being captured. In this university set-ting, however, we have found that onlyone of approximately 150 people (exclud-ing study participants) have a visible Blue-tooth device—far fewer than the numberof people actually carrying Bluetooth-enabled gadgets.

This implies that potential users mustdecide to make their device visible to par-ticipate in the experiment. While somepeople have expressed reluctance to doso because of concerns about power con-sumption and security, this remains aviable option to support the matchingfor many who carry Bluetooth phonesthat are not MIDP compatible.

Privacy implicationsBlueAware, BlueDar, and Serendipity

introduce a significant number of pri-vacy concerns if deployed outside a care-fully controlled experiment with humansubjects’ approval. These privacy issuesmust be reviewed in detail before releas-ing this service to the general public.

BlueAware/BlueDarAll subjects in our experiment have



Figure 4. A small portion of a profilestored on the Serendipity server.

explicitly consented to participate, butwe are also collecting data about devicescarried by people who aren’t directly par-ticipating in the experiment. However,logging a phone’s BTID—for example,00E6D6602B5—does not expose the anindividual’s actual identity. Additionally,because the majority of Bluetooth devicesare sold with default settings set to “notdiscoverable,” we are operating underthe assumption that when a device is con-sciously turned to “visible” mode, theuser knows and accepts that others candetect his or her device.

SerendipityProviding a service that supplies

nearby strangers with a user’s name andpicture is rife with liability and privacyissues. The system must be made toensure that the service never jeopardizesa user’s privacy expectations. The mea-sures we have taken to assuage some ofthese concerns include proximity Webpages, anonymous SMS chat, and theoption of limiting interactions to userswithin a friends-of-friends trust network.Clearly, Serendipity must make as manyprivacy-protecting tools available as pos-sible to maintain user diversity and, mostimportantly, keep everyone safe.

Future applicationsBridging social software introduction

systems with current mobile phone tech-nology enables a diverse suite of applica-tions. Conference participants will be ableto find the right people during an event.Large companies interested in facilitatinginternal collaboration could use Serendip-ity to introduce people who are workingon similar projects but who are not withinone another’s social circles. Individualscould go to a bar and immediately findpeople with common interests.

EnterpriseStatic employee surveys reflect a

severely limited view of an organiza-

tion’s social network. We proposeusing proximity data to infer a socialnetwork’s dynamics. For example,BlueAware data could automaticallybuild a network model of the individ-uals within an organization andthereby quantify the effects of a man-agement intervention.

Additionally, incorporating Serendip-ity into the workplace could instigatesynergistic collaborations by connectingpeople who are working on similarmaterial or by connecting a domain

expert to employees who are workingon a problem in that area. Finally, form-ing groups based on the individuals’inherent communication behavior ratherthan a rigid hierarchy might yield sig-nificant insights to organizationalbehavior. We are in discussions nowwith a large technology company toinstall several BlueDar units within oneof their local campuses and integratethem with an informal knowledge man-agement system.5

DatingThe growth of online dating has

soared over recent years as the stigmaassociated with personal ads diminishes.Serendipity provides users an alternativeto encounters with people they’ve metonly through a computer monitor.Although we need many more users thanour current sample to test Serendipity’sefficacy as a dating tool, we’re talkingwith several online dating companiesabout the possibility of integrating a sim-ilar system in their product lines, whichinvolve millions of active participants.

ConferencesThe need for introduction systems at

events such as large conferences andtrade shows is well established.6 Salespeople can generate proximity Webpages to present their photos and inter-ests, similar to those that publicize theirproducts and expertise. Conference par-ticipants can customize their profiles forconnecting only with individuals whomatch specific interest areas. Our initialdeployment showed Serendipity’s poten-tial as a conference-networking tool.

Beyond SerendipityTechnology-driven societal change is

a hallmark of our era. A new infra-structure of intelligent mobile devicesis influencing culture in unplanned andunprecedented ways. For example,SMS text messaging now generates asignificant fraction of many serviceproviders’ revenues, yet cellular net-work operators originally developedthe protocol as a way for their servicetechnicians to test the network.

Similarly, Serendipity’s main usemight not involve any of the applica-tions described here but rather some-thing less expected. Perhaps by lever-aging trust networks, the system coulddramatically change the trade-offs ofhitchhiking. Coordinating mobile plat-forms with embedded computers incars could facilitate ridesharing andcarpooling.

Human-machine interactions. By equip-ping a physical infrastructure with embed-ded computing and a Bluetooth trans-ceiver, a variation of the Serendipity


Technology-driven societal change is a hallmark of

our era. A new infrastructure of intelligent mobile

devices is influencing culture in unplanned and

unprecedented ways.



system could notify human users ofnearby resources or facilities. For instance,the system could notify users of a nearbypublic restroom. If instead of humanusers, we consider mobile platforms withembedded computers, we can envisionother applications. For instance, busescould wait for passengers on a late con-nection, or delivery vehicles could moreefficiently service pickup and drop-offrequests.

Role-based access control. RBAC is atechnique for assigning user permissionsthat correspond to an organization’sfunctional roles.7 By capturing extensiveuser behavior patterns over time, oursystem has the potential to infer not onlyrelationships among users but also theirpermissions. For example, if two stu-dents working in different labs begin col-laborating at a coffee shop, they couldbe granted constrained entrance accessto each other’s lab.

Public release of Serendipity. Serendipity’sfinal test will be its public launch thissummer, under the SenseSix brand, onwww.sensesix.com. We hope that theapplication will prove to be not onlyrobust but also quite popular within the

realms we have described as well asthose unanticipated.

Mobile phones have becomestandard attire around theglobe. Millions of pocketsand purses hold wireless

transceivers, microphones, and the com-putational horsepower of a desktopcomputer of just a few years ago. Todaythe majority of this processing powergoes unused. This will change, however,when mobile applications shift emphasistoward supporting individual desires toaffiliate with other people who sharetheir interests and goals.

Online dating services and knowledgemanagement systems give us glimpses ofintroduction services, yet the real poten-tial of these new applications will requirean infrastructure of socially curiousmobile devices that untether social soft-ware from the desktop and imbue it intoeveryday life. We believe the mobilephone market is at a critical tippingpoint, where functionality will shift fromthe traditional telephone paradigm to amuch broader socio-centric perspective.We hope that this work represents a stepin that direction.

ACKNOWLEDGMENTSWe are grateful to everyone who participatedthroughout the application development includingTony Pryor, Pedro Yip, Steve Kannan, DoochanHan, Greg Sterndale, Jon Gips, Mat Laibowitz, MaxVan Kleek, and Igor Sylvester. Additionally, wethank members of both Nokia and Nokia Research,particularly Harri Pennanen, Hartti Suomela, SakuHieta, Peter Wakim, Suvi Hiltunen, and Timo Salo-maki.

REFERENCES1. B. Wood et al., Market Share: Mobile Ter-

minals, Worldwide, 4Q04 and 2004, mar-ket report, Gartner Group, Mar. 2004;http://www.gartner.com/DisplayDocu-ment?id=473935.

2. M. Laibowitz, “Parasitic Mobility for Sen-sate Media,” master’s thesis, Media Artsand Sciences, Massachusetts Inst. of Tech-nology, 2004.

3. N. Eagle and A. Pentland, “Interaction Cue-ing Using Wireless Networking and a UserProfile Database,” MIT case no. 10705T,US patent filed 6 May 2004, CombinedShort Radio Network and Cellular Tele-phone Network for Interpersonal Commu-nications.

4. N. Eagle and A. Pentland, “Reality Mining:Sensing Complex Social Systems,” to appearin Personal and Ubiquitous Computing,June 2005.

5. N. Eagle, “Can Serendipity Be Planned?”MIT Sloan Management Rev.. vol. 46, no.1, Fall 2004, pp 10–14.

6. R. Borovoy et al., “Meme Tags and Com-munity Mirrors: Moving from Conferencesto Collaboration,” Proc. 1998 ACM Conf.Computer Supported Cooperative Work,ACM Press, 1998, pp. 159-168.

7. R. Sandhu, D. Ferraiolo, and R. Kuhn, “TheNIST Model for Role Based Access Control:Towards a Unified Standard,” Proc. 5thACM Workshop on Role Based Access Con-trol, ACM Press, 2000, pp. 47-63.

For more information on this or any other comput-ing topic, please visit our Digital Library at http://computer.org/publications/dlib.

the AUTHORS

Nathan Eagle is a PhD candidate in media arts and sciences at the MIT Media Labo-ratory. His research interests include machine learning, mobile computing, andcomplex networks. He has an MS in management science and engineering and anMS in electrical engineering from Stanford University. Contact him at MIT MediaLaboratory, E15-383, 20 Ames St., Cambridge, MA 02139; [email protected].

Alex (Sandy) Pentland is the Toshiba Professor of Media Arts and Sciences at theMIT Media Laboratory and director of the Human Dynamics research group. Hisresearch focuses on wearable computers, health systems, smart environments, andtechnology for developing countries. Contact him at MIT Media Laboratory, E15-383, 20 Ames St., Cambridge, MA 02139; [email protected].

social serendipity: mobilizing social...

Documents