An-Empirical-Study-of-Geographic-User-Activity-Pat

AnEmpiricalStudyofGeographicUserActivityPatternsinFoursquareAnastasi[e-mail][e-mail]@cl.cam.ac.ukMassimilia[e-mail]e-scalestudyofuserbehaviorinFoursquare,conductedonadatasetofabout700thou-sandusersthatspansaperiodofmorethan100days.Weanalyzeusercheckindynamics,demonstratinghowitrevealsmeaningfulspatio-temporalpatternsandof-ferstheopportunitytostudybothusermobilityandur-banspaces.Ouraimistoinformonhowscientificre-searcherscouldutilisedatageneratedinLocation-basedSocialNetworkstoattainadeeperunderstandingofhu-manmobilityandhowdevelopersmaytakeadvantageofsuchsystemstoenhanceapplicationssuchasrecom-mendersystems.IntroductionDuringthepastdecadethewidespreaduseofmobilephoneshasofferedtheopportunitytogaininsightsonhumanmobil-ityatunprecedentedtemporalanduserparticipationscales,usingBluetoothorcellulardata(EagleandPentland2006;Gonzalez,Hidalgo,andBarabasi2008).Theriseofonlinesocialnetworksandserviceshasprovidedanotherusefulsourcewherelocationdataandhumanactivityorrelation-shipsarebeingdescribed.Thelatterhassofarbeenexploitedtobothaddressclas-sicalproblemsarisinginsocialnetworksandproposedirec-tionsfornewapplications.Inparticular,inferringfriend-shipsinasocialnetworkthroughtheexploitationofloca-tioninformationsuchasusergeographicco-occurrenceshasbeensuggested(Crandalletal.2010;Eagle,Pentland,andLazer2009).Theauthorsin(Cranshawetal.2010)improvelinkpredictionbasedonco-occurrencesbytakingintoac-countcontextualinformationsuchastheentropyofaloca-tion,ametricthataccountsforthediversityofuniquevis-itorstoanarea.Thecomplementarytask,i.e.predictingaperson’slocationbasedonwhereherfriendsarehasalsobeentackled(Backstrom,Sun,andMarlow2010).Inthemeantime,OnlineLocation-basedSocialNetworks(LBSN),suchasBrightkite,FoursquareandGowalla,areservicesthatarebuiltuponthenotionofbringingtogethertheplaceswevisitwiththefriendsweconnectto,andduetotheirgrowingpopularitytheypresentapromisingsourceofhumanactivitydata.Copyrightc2011,AssociationfortheAdvancementofArtificialIntelligence().Allrightsreserved.Inthisworkwepresentthefirstlarge-scalestudyofuserbehavioronthemostpopularLBSN,Foursquare.Wehavecollectedapproximately12,000,000usercheckinsoverape-riodof111days,describingthemobilitypatternsofmorethan679,000usersacrossabout3milliongeo-taggedandcategorizedvenues.Ourdatasetcorrespondstoalargesam-pleofuseractivityonFoursquarewithfine-grainedtemporalandspatialinformation.Wepresentananalysisofthegeo-temporaldynamicsofcollectiveuseractivityonFoursquareandshowhowcheckinsprovideameanstouncoverhumandailyandweeklypatterns,urbanneighborhoodpropertiesandrecurrenttransitionsbetweendifferentactivities.OurresultsprovidestrongindicationsthatLBSNspresentexcitingandpromisingresearchopportunities.Theappli-cationpotentialofthisanalysisisremarkableandrangesfrommorepreciselocation/activityrecommendersystemsandtripadvisorstomoregeneralfieldsincludingurbanplan-ningandsocialsciences.FoursquareDataset10010110210310410510610−710−610−510−410−310−210−1100Pr[X≥x]4sq4sqviaTwitter(a)PlaceCheckinsCCDF10010110210310410−510−410−310−210−1100Pr[X≥x](b)UserCheckinsCCDFFigure1:ComplementaryCumulativeDistributionFunc-tionsforthenumberofcheckinsatplaces(left)andthenum-berofcheckinsperuser(right).SinceFoursquareAPIprovidesratelimitedauthorizedac-cess,wehaveresortedtoanotherchannelthroughwhichpublicdataisavailableatlargeamounts:Twittermes-sageswhichcontainFoursquarecheckins.ThroughthepublicstreamofTwittermessages,wehaverecordedap-proximately12milliontimestampedlocationcheckins,gen-eratedby679thousandFoursquareusers,betweenMay,27th2010andSeptember,14th2010.Eachcollectedtweetprovidesapointertothethecorrespondingvenue.Thus,wehaverequestedadditionaldatadirectlyfromFoursquare570ProceedingsoftheFifthInternationalAAAIConferenceonWeblogsandSocialMediaandacquiredthefollowinginformationaboutavenue:ge-ographiccoordinates,category,totalnumberofcheckins,uniquenumberofvisitorsandaddress.Asaresult,wehaveacquiredalargesetwithfine-grainedspatialandtemporaldata.WhilethismethodologyallowsustoacquiredataonlyaboutthesubsetofFoursquareuserswhohaveselectedtosharetheircheckinspubliclyviaTwitter,oursamplerepre-sentsapproximately20%to25%oftheentireFoursquareuserbasewhichamountedintotal3millionusersasofSeptember2010(Techcrunch2010).Asanintroductorysteptoouranalysis,wepresentheresomecommentsabouthowusersshareinformationabouttheirlocations.Thenumberofcheckinsisanindicatorofpopularityforplacesamongusers.InFigure1(a)wereportthecomplementarycumulativedistributionfunction(CCDF)ofthenumberofobservedcheckinsateachloca-tion.Wepresenttwocases:thenumberofcheckinsperlo-cationinourdataset,acquiredviaTwitterover111days,andthenumberoftotalcheckinsreportedonFoursquare(sinceitsinception)foreachvenue.Whileoursamplecontainsonl