资源描述:
《A Framework to Compute Page Importance based on User Behaviors》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、InfRetrieval(2010)13:2245DOI10.1007/s10791-009-9098-8AframeworktocomputepageimportancebasedonuserbehaviorsYutingLiuÆTie-YanLiuÆBinGaoÆZhimingMaÆHangLiReceived:3December2008/Accepted:26May2009/Publishedonline:19June2009ÓSpringerScience+BusinessMedia,LLC2009AbstractThispaperisconcerne
2、dwithaframeworktocomputetheimportanceofwebpagesbyusingrealbrowsingbehaviorsofWebusers.Incontrast,manypreviousapproacheslikePageRankcomputepageimportancethroughtheuseofthehyperlinkgraphoftheWeb.Recently,peoplehaverealizedthatthehyperlinkgraphisincompleteandinaccurateasadatasourceford
3、eterminingpageimportance,andproposedusingtherealbehaviorsofWebusersinstead.Inthispaper,weproposeaformalframeworktocomputepageimportancefromuserbehaviordata(whichcoverssomepreviousworksasspecialcases).First,weuseastochasticprocesstomodelthebrowsingbehaviorsofWebusers.Accordingtothean
4、alysisonhundredsofmillionsofrealrecordsofuserbehaviors,wejustifythattheprocessisactuallyacontinuous-timetime-homogeneousMarkovpro-cess,anditsstationaryprobabilitydistributioncanbeusedasthemeasureofpageimportance.Second,weproposeanumberofwaystoestimateparametersofthestochasticprocess
5、fromrealdata,whichresultinagroupofalgorithmsforpageimportancecom-putation(allreferredtoasBrowseRank).OurexperimentalresultshaveshownthattheproposedalgorithmscanoutperformthebaselinemethodssuchasPageRankandTrust-Rankinseveraltasks,demonstratingtheadvantageofusingourproposedframework.
6、Y.Liu(&)SchoolofScience,BeijingJiaotongUniversity,Beijing,Chinae-mail:liuyt_njtu@hotmail.comT.-Y.LiuB.GaoH.LiMicrosoftResearchAsia,Beijing,ChinaT.-Y.Liue-mail:tyliu@microsoft.comB.Gaoe-mail:bingao@microsoft.comH.Lie-mail:hangli@microsoft.comZ.MaAcademyofMathematicalandSystemsScien
7、ce,CAS,Beijing,Chinae-mail:mazm@amt.ac.cn123InfRetrieval(2010)13:224523KeywordsUserbrowsingprocessContinuous-timetime-homogeneousMarkovprocessStayingtimeBrowseRank1IntroductionPageimportanceisakeyfactorforWebsearch,becauseforcontemporarysearchengines,crawling,indexing,andrankinga
8、reusuallyguidedbyth