Collaborative Filtering with Temporal Dynamics

Collaborative Filtering with Temporal Dynamics

ID:39910709

大小:700.57 KB

页数:9页

时间:2019-07-14

Collaborative Filtering with Temporal Dynamics _第1页
Collaborative Filtering with Temporal Dynamics _第2页
Collaborative Filtering with Temporal Dynamics _第3页
Collaborative Filtering with Temporal Dynamics _第4页
Collaborative Filtering with Temporal Dynamics _第5页
资源描述:

《Collaborative Filtering with Temporal Dynamics 》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、CollaborativeFilteringwithTemporalDynamicsYehudaKorenYahoo!Research,Haifa,Israelyehuda@yahoo-inc.comABSTRACTtemporaryeffectsthathaveverylowimpactonfuturebehavior,Customerpreferencesforproductsaredriftingovertime.Productwhilecapturinglonger-termtrendsthatreflecttheinhere

2、ntnatureperceptionandpopularityareconstantlychangingasnewselec-ofthedata.Thisledtomanyworksontheproblem,whichisalsotionemerges.Similarly,customerinclinationsareevolving,lead-widelyknownasconceptdrift;seee.g.[15,25].ingthemtoeverredefinetheirtaste.Thus,modelingtemporaldy

3、-Modelingtemporalchangesincustomerpreferencesbringsuniquenamicsshouldbeakeywhendesigningrecommendersystemsorchallenges.Onekindofconceptdriftinthissetupistheemergencegeneralcustomerpreferencemodels.However,thisraisesuniqueofnewproductsorservicesthatchangethefocusofcusto

4、mers.Re-challenges.Withintheeco-systemintersectingmultipleproductslatedtothisareseasonalchanges,orspecificholidays,whichleadandcustomers,manydifferentcharacteristicsareshiftingsimulta-tocharacteristicshoppingpatterns.Allthosechangesinfluencetheneously,whilemanyoftheminflu

5、enceeachotherandoftenthosewholepopulation,andarewithintherealmoftraditionalstudiesonshiftsaredelicateandassociatedwithafewdatainstances.Thisconceptdrift.However,manyofthechangesinuserbehavioraredistinguishestheproblemfromconceptdriftexplorations,wheredrivenbylocalizedf

6、actors.Forexample,achangeinthefamilymostlyasingleconceptistracked.Classicaltime-windoworinstance-structurecandrasticallychangeshoppingpatterns.Likewise,indi-decayapproachescannotwork,astheylosetoomuchsignalwhenvidualsgraduallychangetheirtasteinmoviesandmusic.Allthosedi

7、scardingdatainstances.Amoresensitiveapproachisrequired,changescannotbecapturedbymethodsthatseekaglobalconceptwhichcanmakebetterdistinctionsbetweentransienteffectsanddrift.Instead,foreachcustomerwearelookingatdifferenttypeslongtermpatterns.Theparadigmweofferiscreatingam

8、odelofconceptdrifts,eachoccursatadistincttimeframeandisdriventrackingthetimechangingbehaviorthroughoutthelifespanofth

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。