欢迎来到天天文库
浏览记录
ID:7290214
大小:160.37 KB
页数:11页
时间:2018-02-10
《manipulation-resistant reputations system》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、Manipulation-ResistantReputationsUsingHittingTimeJohnHopcroftDanielSheldonCornellUniversityCornellUniversityjeh@cs.cornell.edudsheldon@cs.cornell.eduAbstractPopularreputationsystemsforlinkednetworkscanbemanipulatedbyspammerswhostrategicallyplacelinks.Thereputationofnodevisinterpretedasthewo
2、rld’sopinionofv’simportance.InPageRank[4],v’sownopinioncanbeseentohaveconsiderableinfluenceonherreputation,wherevexpressesahighopinionofherselfbyparticipatinginshortdirectedcycles.Incontrast,weshowthatexpectedhittingtime—thetimetoreachvinarandomwalk—measuresessentiallythesamequantityasPageRan
3、k,butexcludesv’sopinion.Wemakethesenotionsprecise,andshowthatareputationsystembasedonhittingtimeresiststamperingbyindividualsorgroupswhostrategicallyplaceoutlinks.Wealsopresentanalgorithmtoefficientlycomputehittingtimeforallnodesinamassivegraph;conventionalalgorithmsdonotscaleadequately.1Intr
4、oductionReputationandrankingsystemsareanessentialpartofwebsearchande-commerce.Thegeneralideaisthatthereputationofoneparticipantisdeterminedbytheendorsementsofothers;forexam-ple,onewebpageendorsesanotherbylinkingtoit.However,notallparticipantsarehonorable—e.g.,spammerswilldotheirbesttomanipul
5、ateasearchengine’srankings.Anaturalrequirementforareputationsystemisthatindividualsshouldnotbeabletoimprovetheirownreputationus-ingsimpleself-endorsementstrategies,likeparticipatinginshortcyclestoboostPageRank.SincePageRankenjoysmanyniceproperties,itisinstructivetoseewherethingsgowrong.LetG=
6、(V;E)beadirectedgraph(e.g,theweb).PageRankassignsascore(v)toeachnodev,whereisdefinedtobethestationarydistributionofarandomwalkonG,givingthepleasinginterpretationthatthescoreofpagevisthefractionoftimeawebsurferspendsthereifsherandomlyfollowslinksforever.Fortechnicalreasons,therandomwalkismod
7、ifiedtorestartineachstepwithprobability,jumpingtoapagechosenuniformlyatrandom.Thisensuresthatexistsandisefficienttocompute.Thenawell-knownfactaboutMarkovchains[1]saysthat1=(v)isequaltotheexpectedreturntimeofv,thenumberofstepsittakesarandomwalkstart
此文档下载收益归作者所有