基于主題模型的社會化推薦系統(tǒng)研究.pdf

1、ADissertationSubmittedtoZhejiangUniversityfortheDegreeofMasterofEngineering1一一⑧TITLE：ResearchofaSocia—l—RecommendationSystembasedonTopicModelsAuthor：Supervisor：Subject：College：ComputerAppllcatmn1echnology—————————1r—————

2、—————。。———1r—，r_一一一。SubmittedDate：20l3／3／18浙江大學(xué)碩士學(xué)位論文AbstractAbstractWiththeexplosivegrowthofonlinesocialnetworks，thesocialnetworkbasedapproachestorecommendationhavebeenwellstudiedanddevelopedIntherealworld，usersofsocial

3、networkalwayschooseinformationswithrespecttotheircontentButmostsocialrecommendersystemsarebuiltonthecombinationofcollaborativefilteringandsocialtrustpropagationmodel，withoutconcerningthetextcontentofitemsToincreasetheacc

4、uracyofsocialrecommendation，weintroducetopicmodelsintosocialrecommendation，takingthetopicsofitemsasfactorsofusers’ratingsWefirstanalysethetechniquesoftopicmodelingandsocialrecommendingBasedonourstudies，weproposeatrust—ba

5、sedtopicmodelnamedTBLDA(TrustBasedfactorizationmodelofLatentDirichletAllocation)forrecommendationsystemsInmodelfittingprocedure，wefactorizetheratingmatrixtoobtainlatentfeaturefactors，andconsidertheinfluenceofsocialconnec

6、tionsonpersonalaffinityFittedmodelcanbeusedtopredictunknownratingsandproviderecommendationforusersInordertoformmorepredictiveitemfactors，weadoptLDAtopicmodeltoinferlatenttopicsofitemsThetrainingphaseofLDAtakesobservedrat

7、ingsasbasisofposteriorprobabilityThissupervisedlearningmethodwillachievebetterpredictionsforparcticalpurposeofrecommendationThenweconductexperimentsonadatasetcrawledfromWeibocomTheexperimentalresultsdemonstratethatourapp