資源描述:
《learning to detect vandalism in social content systems a study on wikipedia》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在工程資料-天天文庫(kù)。
1、LearningtoDetectVandalisminSocialContentSystems:AStudyonWikipediaVandalismDetectioninWikipediaSaraJavanmardi,DavidW.McDonald,RichCaruana,SholehForouzan,andCristinaV.LopesAbstractAchallengefacingusergeneratedcontentsystemsisvandalism,i.e.editsthatdamagecontentquality.Thehighvisibilityandeasy
2、accesstosocialnetworksmakesthempopulartargetsforvandals.Detectingandremovingvandalismiscrit-icalfortheseusergeneratedcontentsystems.Becausevandalismcantakemanyforms,therearemanydifferentkindsoffeaturesthatarepotentiallyusefulforde-tectingit.Thecomplexnatureofvandalism,andthelargenumberofpot
3、entialfea-tures,makevandalismdetectiondif?cultandtimeconsumingforhumaneditors.Machinelearningtechniquesholdpromisefordevelopingaccurate,tunable,andmaintainablemodelsthatcanbeincorporatedintovandalismdetectiontools.Wedescribeamethodfortrainingclassi?ersforvandalismdetectionthatyieldsclassi-?
4、ersthataremoreaccurateonthePAN2010corpusthanotherspreviouslydevel-oped.Becauseofthehighturnaroundinsocialnetworksystems,itisimportantforvandalismdetectiontoolstoruninreal-time.Tothisaim,weusefeatureselectionto?ndtheminimalsetoffeaturesconsistentwithhighaccuracy.Inaddition,becausesomefeature
5、saremorecostlytocomputethanothers,weusecost-sensitivefeatureselectiontoreducethetotalcomputationalcostofexecutingourmodels.Inadditiontothefeaturespreviouslyusedforspamdetection,weintroducenewfeaturesbasedonuseractionhistories.Theuserhistoryfeaturescontributesigni?cantlytoclassi-?erperforman
6、ce.Theapproachweuseisgeneralandcaneasilybeappliedtootherusergeneratedcontentsystems.S.Javanmardi(B)UniversityofCalifornia,IrvineDonaldBrenHall5042,Irvine,CA92697-3440,USAe-mail:sjavanma@ics.uci.eduD.W.McDonaldTheInformationSchool,UniversityofWashington,Washington,WA,USAR.CaruanaMicrosoftRes
7、earch,Redmond,WA,USAS.Forouzan·C.V.LopesBrenSchoolofInformationandComputerSciences,UniversityofCalifornia,Irvine,CA,USAT.?zyeretal.(eds.),MiningSocialNetworksandSecurityInformatics,203LectureNotesinSocialNetworks,DOI10.1007/978-94-007-6359-3_11,?Springer