电子商务个性化推荐系统中协同过滤算法的研究

南京航空航天大学硕士学位论文电子商务个性化推荐系统中协同过滤算法的研究姓名：高翔申请学位级别：硕士专业：管理科学与工程指导教师：马静2011-03IICF-ADICF-ADICF-ADICF-ADIIAbstractWiththepopularityoftheInternetande-commerceapplication,itisdifficultforthemtofindtheirneededproductswithinamassofproductinformation.Therefore,therecommendationsystemine-commercecameintobeing.Recommendationsystemswereproposedtoprovidethetargetuserwithinformationtheyinterestedinbasedontheexistinginformation,souserscanbemoreconvenienttofindtheinformationtheyneed.Collaborativefilteringisoneofthemostwidelyusedandsuccessfulmethodsforrecommendation.It’ssignificanttostudyit.Weintroducedthebasicknowledgeofrecommendationsystemandmainrecommendationskilldetailedlyinthisthesis,andanalyzethemostpopularuser-basedcollaborativefilteringandtheitem-basedcollaborativefilteringinthecollaborativefilteringalgorithms,thenleadthebottlenecksproblemofcollaborativefiltering:datasparsity,coldstartandscalability,summarizingsomesolutions.Weproposedtwoimprovedmethodsforsparseproblemfromtwodifferentparts,andbringtheimprovedcollaborativefiltering.Oneofthemethodsistominetheassociationoftheseitemsthroughtheratingmatrix,thencomplementthematrixwiththeseassociations;anotheroneisbringingtheconceptof“domain”andcalculatingthesimilarityoftwousersinthedomain.Itnotonlyensurethatthereliabilityofthespacebutalsoreducethealgorithmcomplexityandtimecomplexity.Wetheoreticallydetailedanalyzethenewmethodsandprovetheirfeasibility.Thentheexperimentalresultsthatthenewmethodisimplementedwiththebenchmarkexperimentaldatasetaregiven,theperformancebetweenthenewmethodsandtheoldmethodsiscomparedandanalyzed.Theexperimentsshowthattheimprovedmethodscanalleviatetheproblemofsparsityeffectivelyandscalability,improvethequalityofrecommendationandefficiency.Keywords:RecommendationSystems,CollaborativeFiltering,Association,DataSparsityV1.1............................................................61.2..............................................................................61.3...............................................................................................81.4WebMate...........................................................................................91.5...............................................................101.6CBFCF.................................................................................111.7..............................................................................................111.8Tapestry............................................................................................121.9GroupLens........................................................................................132.1..............................................................................................202.2Fab...........................................................................................252.3.......................................................................................263.1ICF-AD...........................................................................394.1MAE.......................................................................................................434.2..................................................................................................431.1......................................................................131.2....................................................................................142.1..............................................................................................182.2...............................................................222.3...............................................................233.1...........................................................................................333.2.....................................................................................................333.3.......................................................................................334.1................................................................................................................404.2.........................................................................................................404.3.........................................................................................................414.4.....................................................................................................424.5............................................................................................................4211.1Internet[1]123Alibaba[2]2NielsenNetRating3547%52%JupiterCommunications503%-9%[3]CollaborativeFilteringCF[4]sparsitycold-startscalability31.21[5][6][7][8][9]2(serendipitousrecommendations)1.3(PersonalizedRecommenderSystems)Resnick&Varian41997[10]121PersonalizedRecommendation2socialrecommendation3itemrecommendation[11]1235explicitdataimplicitdata1[12]2[12]Claypool[13]CuriousBrowser[14]Nichols[15]13/6[16]behaviorbasedattitudebased1.11.1Al-Shamri[17]demographicaldata1.21.271[18]RelationDatabaseSQL1.381.3TF-IDFtermfrequency/inversedocumentfrequ