CLR:ACollaborativeLocationRecommendation
FrameworkbasedonCo-Clustering
KennethWai-TingLeungkwtleung@cse.ust.hk
ABSTRACT
GPSdatatrackedonmobiledevicescontainsrichinformationabouthumanactivitiesandpreferences.Inthispaper,GPSdataisusedinlocation-basedservices(LBSs)toprovidecollaborativelocationrecommendations.WeobservethatmostexistingLBSsprovidelocationrecommendationsbyclusteringtheUser-Locationmatrix.SincetheUser-LocationmatrixcreatedbasedonGPSdataishuge,therearetwomajorproblemswiththesemethods.First,thenum-berofsimilarlocationsthatneedtobeconsideredincomputingtherecommendationscanbenumerous.Asaresult,theidenti cationoftrulyrelevantlocationsfromnumerouscandidatesischalleng-ing.Second,theclusteringprocessonlargematrixistimecon-suming.Thus,whennewGPSdataarrives,completere-clusteringofthewholematrixisinfeasible.Totacklethesetwoproblems,weproposetheCollaborativeLocationRecommendation(CLR)frameworkforlocationrecommendation.Byconsideringactivi-ties(i.e.,temporalpreferences)anddifferentuserclasses(i.e.,Pat-ternUsers,NormalUsers,andTravelers)intherecommendationprocess,CLRiscapableofgeneratingmorepreciseandre nedrecommendationstotheuserscomparedtotheexistingmethods.Moreover,CLRemploysadynamicclusteringalgorithmCADCtoclusterthetrajectorydataintogroupsofsimilarusers,similarac-tivitiesandsimilarlocationsef cientlybysupportingincrementalupdateofthegroupswhennewGPStrajectorydataarrives.WeevaluateCLRwithareal-worldGPSdataset,andcon rmthattheCLRframeworkprovidesmoreaccuratelocationrecommendationscomparedtotheexistingmethods.
HongKongUniversityofScienceandTechnologyHongKongUniversityofScienceandTechnology
DikLunLee
dlee@cse.ust.hk
ThePennsylvaniaState
University
Wang-ChienLee
wlee@cse.psu.edu
1.INTRODUCTION
Oneofthemajorfunctionsofalocation-basedservice(LBS)istorecommendinterestinglocationstotheusers.MostexistingLBSs[2]recommendallpointsofinterest(POIs)thatareneartheusers.However,thenumberofPOIsthatareclosetotheuser’scurrentposition(calledtheactivelocationinthispaper)canbenu-merous.Thus,itisdif cultforausertopickthemostrelevantsuggestionsfromthelargenumberofrecommendations.Inor-dertoprovidemoreaccuratelocationsuggestionsaccordingtotheusers’interests,manyexistingLBSs[1],[9],[10],[15]providelo-cationrecommendationsby rstclusteringtheUser-Locationma-trix,whichrepresentsthelocationsvisitedbyeachuser,andthenmakinglocationrecommendationsbasedontheuserandlocationclusters.However,amajorproblemwiththesemethodsisthattheUser-LocationmatrixcreatedbasedonGPSdataishuge,sothenumberofrecommendationsisstillverylargeandtheproblemofselectingthetrulyrelevantlocationsremainschallenging.
Totacklethisproblem,weproposethecollaborativelocationrecommendation(CLR)frameworkforlocationrecommendation.CLRcollectsusers’GPStrajectorydatatodiscoverasetofpointsofinterest(POIs)ascandidatelocations.Insteadofmerelycon-sideringusersandlocationsrepresentedinaUser-Locationmatrix,weintroducetheCommunityLocationModel(CLM),whichincor-poratesactivitiesinadditiontousersandlocations.CLMcapturestherelationsbetweenthethreeimportantentities,namely,users,ac-tivitiesandlocations,inGPStrajectorydatawithaUser-Activity-Locationtripartitedatastructure,whichwecallCLMgraphforsimplicity.Wearguethatactivitiesareimportantbecausemostusersvisitlocationswiththeintentiontoperformsomeactivities(e.g.,gotoschooltostudy,gotoof cetowork,gotoacinematowatchmovie).AnoveltyofCLMisitsabilitytoderivefromGPStrajectoriesusers’temporalpreferencesthroughtheiractivi-ties.CLMexplorestheuniquepropertiesamongusers,activities,andlocationstomakeeffectivelocationrecommendations.Weob-servethat(1)locationsexhibitspatialpropertiesastheycontainge-ographicalinformation;(2)activitiesexhibittemporal-spatialprop-ertiesastheyrepresenttemporalsequencesofvisitedlocations;and(3)usersexhibitlong-termspatialpropertiesasdifferentuserstendtovisitdifferentlocationsaccordingtotheirlongtermhabitsandgeographicallimits.Wearguethat,bycapturingtemporal-spatialinformationinuseractivities,theproposedCLMismoreeffectiveindiscoveringsimilarlocationscomparingtoexistingtechniquesthatarebasedontheUser-Locationmatrix.
InCLR,aco-clusteringalgorithmCADCisemployedtoclustertheCLMgraphinordertoobtainhigh-qualitylocationrecommen-dations.ThereareacoupleofadvantagesofapplyingCADContheCLMgraph.First,byexploringtherelationsbetweenusers,activitiesandlocationsmodeledinCLM,CADCcaniteratively
CategoriesandSubjectDescriptors
H.2.8[DatabaseManagement]:DatabaseApplications—DataMin-ing;H.3.3[InformationStorageandRetrieval]:InformationSearchandRetrieval—Clustering,InformationFiltering
GeneralTerms
Algorithms,Experimentation
Keywords
Co-Clustering,CollaborativeFiltering,LocationRecommendation
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforpro torcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthe rstpage.Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspeci cpermissionand/orafee.
SIGIR’11,July24–28,2011,Beijing,China.
Copyright2011ACM978-1-4503-0757-4/11/07...$10.00.
recommendation technology, a Clothing Recommend System based on radio frequency...Merugu, A Scalable Collaborative Filtering Framework Based On Co-clustering, ...
WWW COLLABORATIVE RECOMMENDATION BASED ON RELIABILITY_专业资料 暂无评价
User-based collaborative filtering ? Item-based ...Hierarchical clustering GPS Logs of User 1 GPS ...Framework Friend and Location Recommendation Similar ...
(10):2403-2411.Zhang Guangwei,Li Deyi,Li Peng,et al.A collaborative filtering recommendation algorithm based on cloud model〔 J〕. Jour darion algorithms...
Item-based Collaborative Filtering Recommendation Algorithms_IT/计算机_专业资料。this paper analyzed different item-based recommendation generation algorithms. We lo...
(山东建筑大学 计算机科学与技术学院,山东 济南 100083) 518055) 250101) A Collaborative Filtering Recommendation Algorithm Based on Cloud Model ZHANG Guang-Wei ...
WWW COLLABORATIVE R... 3人阅读 4页 2.00 User Interfaces of Cur... 8人阅读 6页 2.00 A Process Meta- Model ... 1人阅读 14页 2.00...
Collaborative-filtering-based recommendation systems ...Genetic algorithm first completes clustering for ...T. (1999). An algorithmic framework for ...
While recommendation system based on collaborative filtering could learn users’ new interest by users’ similar action, but it suffers from cold start [3~...
Clustering those session vectors Monitoring of ...Main Methods Collaborative filtering Iterative ...Recommendation -> Personalized Algorithms Accuracy ...
热门文档