Flexible mixture model for collaborative filtering.pdf

《Flexible mixture model for collaborative filtering.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、FlexibleMixtureModelforCollaborativeFilteringLuoSiLSI@CS.CMU.EDURongJinRONG@CS.CMU.EDUSchoolofComputerScience,CarnegieMellonUniversity,Pittsburgh,PA15232USAAbstractusersinthetrainingdatabasesimilartothetestuserandthen,predictthetestuser’sratingsbasedontheThispaperpresen

2、tsaflexiblemixturemodelcorrespondingratingsofthesesimilarusers.Onthe(FMM)forcollaborativefiltering.FMMextendscontrary,model-basedalgorithmsbuildmodelsthatareexistingpartitioning/clusteringalgorithmsforabletoexplainthetrainingexampleswellandpredictthecollaborativefilteri

3、ngbyclusteringbothusersratingsoftestusersusingtheestimatedmodels.Bothofanditemstogethersimultaneouslywithoutthememory-basedalgorithmsandthemodel-basedassumingthateachuseranditemshouldonlyalgorithmshavetheiradvantagesanddisadvantages.belongtoasinglecluster.Furthermore,wi

4、ththeMemory-basedalgorithmshavemuchlessoff-lineintroductionof‘preference’nodes,theproposedcomputationcostswhilethemodel-basedalgorithmsmayframeworkisabletoexplicitlymodelhowusershavelesson-linecomputationcosts.rateitems,whichcanvarydramatically,evenamongtheuserswithsimi

5、lartastesonitems.Thoughmemory-basedandmodel-basedapproachesEmpiricalstudyovertwodatasetsofmoviedifferfromeachotherinmanyaspects,bothofthemratingshasshownthatournewalgorithmassumethatuserswithsimilartastesshouldrateitemsoutperformsfiveothercollaborativefilteringsimilarly

6、andthereforetheideaofclusteringisusedinalgorithmssubstantially.bothapproacheseitherexplicitlyorimplicitly.Formemory-basedapproaches,traininguserssimilartothetestuseraregroupedtogetherandtheirratingsare1.Introductioncombinedtopredictratingsforthetestuser.Meanwhile,model-

7、basedapproachesclusteritemsand/ortrainingTherapidgrowthoftheinformationontheInternetusersintoclassesexplicitlyandpredictratingsofatestdemandsintelligentinformationagentthatcansiftuserbysimplyusingtheratingsofclassesthatfitinbestthroughalltheavailableinformationandfindou

8、tthewiththetestuserand/oritemstoberated.Thus,howtomostvaluabletous.Theseintelligentsystemscanbeclusterusersand