Time-dependent Models in Collaborative Filtering based Recommender System-xlvector.pdf

《Time-dependent Models in Collaborative Filtering based Recommender System-xlvector.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、2009IEEE/WIC/ACMInternationalConferenceonWebIntelligenceandIntelligentAgentTechnology-Workshops2009IEEE/WIC/ACMInternationalJointConferenceonWebIntelligenceandIntelligentAgentTechnologyTime-dependentModelsinCollaborativeFilteringbasedRecommenderSystemLiangXiangQingYangChineseAc

2、ademyofSciencesChineseAcademyofSciencesInstituteofAutomationInstituteofAutomationNationalLaboratoryofPatternRecognitionNationalLaboratoryofPatternRecognitionBeijing,ChinaBeijing,Chinaxlvector@gmail.comqyang@nlpr.ia.ac.cnAbstract—Inrecentyears,timeinformationismoreandanditem-ite

3、msimilarity.Neighborhoodmethodsareeasytomoreimportantincollaborativefiltering(CF)basedrecom-implementandwidelyusedinmanyrecommendersystemsmendersystembecausemanysystemshavecollectedrating[2][5][6].dataforalongtime,andtimeeffectsinuserpreferenceisAnothertypeofmethodoftenusedinCFi

4、smatrixstronger.Inthispaper,wefocusonmodelingtimeeffectsinCFandanalyzehowtemporalfeaturesinfluenceCF.Therearefactorization,whichisalsoknownaslatentclassmodelfourmaintypesoftimeeffectsinCF:(1)timebias,theinterest[10][11].Itskeyideaisusingalowrankmatrixtoap-ofwholesocietychangeswi

5、thtime;(2)userbiasshifting,proximaterealratingmatrix.SingularValueDecompositionausermaychangehis/herratinghabitovertime;(3)item(SVD)[12][13][14]isoftenusedtocalculatelowrankratingbiasshifting,thepopularityofitemschangeswithtime;(4)matrix,thusthesefactorizedmethodsareoftencalled

6、SVD.userpreferenceshifting,ausermaychangehis/herattitudetosometypesofitems.Inthiswork,thesefourtimeeffectsareManyresearchesaboutCFshowfactorizationbasedmethodsusedbyfactorizedmodel,whichiscalledTimeSVD.Moreover,canproducemoreaccuratepredictionsthanneighborhoodmanyothertimeeffec

7、tsareusedbysimplemethods.Ourtime-basedmethods.dependentmodelsaretestedonNetflixdatafromNov.1999toNowadays,manyrecommendersystemhavecollectedDec.2005.ExperimentalresultsshowthatpredictionaccuracyuserpreferencedataforalongtimeandtimeinformationisinCFcanbeimprovedsignificantlybyusin

8、gtimeinformation.moreandmoreimportantinmakingrecommend