文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 互联网 > Personalizing information appliances-- a multi-agent framework for TV programme recommendations
侵权投诉

Personalizing information appliances-- a multi-agent framework for TV programme recommendations

国外计算机科学类论文,推荐系统

332W.-P.Lee,T.-H.Yang/ExpertSystemswithApplications25(2003)331–341

environmentsuchastelevision,anidealsolutiontotheuseristohaveapersonalidenti er(liketheSIMcardinthemobilephone)thatcanrecordhismostrecentpreferencesandworkasa lterforchannelselection.Therefore,wheneverauserequipshisidenti erintoaspecialdevice(e.g.remotecontrol)toturnonthetelevision,theprogrammesavailableatthattimearesortedautomaticallyandpresentedtohimbasedonhispreferences.Inotherwords,the rstprogrammepresentedtoauserispresumedtobetheonehelikesthemostamongtheprogrammescurrentlybroadcasted;thenextoneisthesecondpreferred,andsoon.Inthisway,theaudiencecanimmediatelyaccesstheprogrammesheisinterestedinandthensavemuchofthetimespentonreadingtheinformationaboutthecontentsorbrowsingthechannelstodecidewhattowatch.

Inthispaper,wedescribehowweadoptanagent-basedmethodologytodevelopaprototypesystemforpersonalizedTVprogrammerecommendations.Twotypesofprogrammesareconsidered:oneisthe lm,andtheother,thenews.Todealwithdifferentprogrammes,oursystemincludes veagents:a lmagenttoimplicitlyobserveauser’ssur ngbehaviorandthenrecordwhathewatches,ortogatherauser’spreferencesinanexplicitway;anewsagenttocollectInternetnewsforuser’sidenti cation;alearningagenttoanalyzeandmodelauser’sinterests;acollaborationagenttolookforotheruserswithsimilartastesforsharingtheinformationcollected;andanadaptationagenttoevaluatetheperformanceofthesystemandtoadapttotheuser’snewlychangedinterests.Theexperimentalresultsshowthatourlearningagentcanlearnauser’sinterestsinrealtimeandgivesensiblerecommendations.Inaddition,thek-nearestneighbormethodisimplementedasabaselineforperformancecomparison.Furthermore,wealsoshowthatfordifferentsituationsinwhichauserchoosesaprogrammewithlowerpriorityratherthanthetoponesuggestedbythesystem,thelearningagentcanberesumedaccordinglyanditcontinuouslylearnsthemostup-to-datepreferencesfortheuser.

2.Background

Ingeneral,therearetwokindsofapproachesindevelopingarecommendersystem:personalized(content-based)andsocialized(collaborative)ones.Inaperson-alizedsystem,auser’spersonalinformationis rstcollected,thenthesystemreasonstheuser’spreferencesbyanalyzingandmodelingthepersonalinformationavailable.Theensuinguserpro leisrecordedbythesystemforfurtherrecommendations.Tocollectinformation,thesystemcanaskausertomanuallyindicatehisinterestsbygivingsomeexampleswithpersonalratings.Alternativelythesystemcanimplicitly

observeandrecordtheuser’soperatingbehavior,andthenanalyzesthecontenttheuserhasreadinordertobuildamodeltodistinguishbetweenwhathelikesandwhathedoesnot.

Themostimportantissueinthepersonalizationworkistoconstructacomputationalmodelforeachindividualusertopredicthispreferencesforincominginformation.Withtheusermodelworkingasa lter,onlytheinformationconsideredasuser-interestedcanreachtheenduser.Therefore,theworkofrecommendationcanberegardedasclassi cation:usingtheinformationalreadyknowntobuildamodeltopredicttheeventsunknown.Themodelmaybeasetofrules,adecisiontree,orasetofnumericalnumbersrepresentingthecorrespondingweightsforsomespeci cfeatures.Somemachinelearningapproacheshavebeenappliedtoconstructusers’pro les.InBalabanovicandShoham(1997),astatistic-basedapproacht df(termfrequencyandinversedocumentfrequency(Salton,1989))wasusedtobuildauser’spro letorecommendWebpages.InJoachims,Freitag,andMitchell(1997),areinforcementlearningmethodwasemployedforWebpagerecommendations,andinMooneyandRoy(2000),forbookrecommen-dations.Inthiswork,weuseadecisiontree-basedapproachtodevelopalearningagenttomodelauser’sinterestsforTVprogrammeselections,andtoexploredifferentprogrammefeaturesforbetterrecommendations.Insteadofbuildingapersonalmodeltopredicthowauserlikesanitem,acollaborativesystemrecommendsitemstoaspeci cuseraccordingtotheevaluationsfromotheruserswithsimilartastes.Inotherwords,thesystemdoesnotanalyzewhatauserlikesbuttakingtheopinionsofothers.Insuchasystem,thesimilarityoftheusersismeasuredbycertaincorrelationcriterion,andthenak-nearestneighbormethodisusedto ndsomeuserswithsimilarinterestsforaspeci cuser.Thepredictionofanunknownitemforauseristhusbasedonthecombinationoftheratingsofhisnearestneighbors.Goodetal.(1999)andShardanandandMaes(1995)presentsystemsofthiskind.

Somehybridmethodsareproposedtoovercometheproblemswiththeabovetwokindsofrecommendersystems,forexample,thecombinedtechniquesforpersonalizationandcollaborationusedinBalabanovicandShoham(1997),Basu,Hirsh,andCohen(1998),andSmythandCotter(2000).InSmythandCotter(2000),apersonalpro leiscompiledfromtheTVguideforprogrammerecommendations.Ourworkpresentedherediffersfromtheirsnotonlyinthelearningapproachused,butalsointhesituationsoftwodifferenttypesofprogrammesbroadcasted(i.e. lmsandnews).Mostimportantly,ourworkincludesaspeciallydesignedmechanismforsystemadaptation,withwhichauser’spro lecanbere-learnttocorrectthepredictionerrorsandtomeettheuser’smostrecentpreferences.

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email