资源描述：
《A Contextual-Bandit Approach to语境强盗方法 个性化新闻推荐》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、WWW2010•FullPaperApril26-30•Raleigh•NC•USAAContextual-BanditApproachtoPersonalizedNewsArticleRecommendation∗LihongLi†,WeiChu†,JohnLangford‡RobertE.Schapire+†‡+DeptofComputerScienceYahoo!LabsYahoo!Labslihong,chuwei@yahoo-jl@yahoo-inc.comPrincetonUniversityinc.comschapire@cs.princeton.eduABSTRACTserv
2、icevendorsacquireandmaintainalargeamountofcontentintheirrepository,forinstance,forfilteringnewsarticles[14]orforPersonalizedwebservicesstrivetoadapttheirservices(advertise-thedisplayofadvertisements[5].Moreover,thecontentofsuchaments,newsarticles,etc.)toindividualusersbymakinguseofweb-servicereposit
3、orychangesdynamically,undergoingfrequentbothcontentanduserinformation.Despiteafewrecentadvances,insertionsanddeletions.Insuchasetting,itiscrucialtoquicklythisproblemremainschallengingforatleasttworeasons.First,identifyinterestingcontentforusers.Forinstance,anewsfilterwebserviceisfeaturedwithdynamica
4、llychangingpoolsofcon-mustpromptlyidentifythepopularityofbreakingnews,whilealsotent,renderingtraditionalcollaborativefilteringmethodsinappli-adaptingtothefadingvalueofexisting,agingnewsstories.cable.Second,thescaleofmostwebservicesofpracticalinterestItisgenerallydifficulttomodelpopularityandtemporalc
5、hangescallsforsolutionsthatarebothfastinlearningandcomputation.basedsolelyoncontentinformation.Inpractice,weusuallyex-Inthiswork,wemodelpersonalizedrecommendationofnewsploretheunknownbycollectingconsumers’feedbackinrealtimearticlesasacontextualbanditproblem,aprincipledapproachintoevaluatethepopular
6、ityofnewcontentwhilemonitoringchangeswhichalearningalgorithmsequentiallyselectsarticlestoserveinitsvalue[3].Forinstance,asmallamountoftrafficcanbedes-usersbasedoncontextualinformationabouttheusersandarticles,ignatedforsuchexploration.Basedontheusers’response(suchwhilesimultaneouslyadaptingitsarticle
7、-selectionstrategybasedasclicks)torandomlyselectedcontentonthissmallsliceoftraf-onuser-clickfeedbacktomaximizetotaluserclicks.fic,themostpopularcontentcanbeidentifiedandexploitedontheThecontributionsofthiswor
