资源描述：
《a probabilistic reputation model based on transaction ratings》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、AProbabilisticReputationModelbasedonTransactionRatings∗Franc¸oisFouss†ManagementDepartment–LSMFacult´esUniversitairesCatholiquesdeMons(FUCaM),BelgiumEmail:francois.fouss@fucam.ac.be,Tel:+326532.32.16YoussefAchbany&MarcoSaerensInformationSystemsResearchUnit(ISYS)–LS
2、MUniversit´ecatholiquedeLouvain(UCL),BelgiumEmail:{youssef.achbany,marco.saerens}@ucLouvain.beSeptember22,2009AbstractThisworkintroducesaprobabilisticmodelofreputationallowingtocomputereputa-tionscoresascloseaspossibletotheirintrinsicvalue,accordingtothemodel.Itisb
3、asedonthefollowing,natural,consumer-providerinteractionmodel.Consumersareassumedtoorderitemsfromproviders,whoeachhassomeintrinsic,latent,“qualityofservice”score.Inthebasicmodel,theproviderssupplytheitemswithaqualityfollowinganormallaw,centeredontheirintrinsic“quali
4、tyofservice”.Theconsumers,afterthereceptionandtheinspectionoftheitem,rateitaccordingtoalinearfunctionofitsquality–astandardregressionmodel.Thisregressionmodelaccountsforthebiasoftheconsumerinprovidingratingsaswellashisreactivitytowardschangesinitemquality.Moreover,
5、theconstancyoftheproviderinsupplyinganequalqualitylevelwhendeliveringtheitemsisestimatedbythestandarddeviationofhisnormallawofitemqualitygeneration.Symmetrically,theconsistencyoftheconsumerinprovidingsimilarratingsforagivenqualityisquantifiedbythestandarddeviationof
6、hisnormallawofratingsgeneration.Twoextensionsofthisbasicmodelareconsideredaswell:amodelaccountingfortruncationoftheratingsandaBayesianmodelassumingapriordistributionontheparameters.Expectation-maximizationalgorithms,allowingtoestimatetheparametersbasedontheratings,
7、aredevelopedforallthemodels.Theexperimentssuggestthatthesemodelsareabletoextractusefulinformationfromtheratings,arerobusttowardsadversebehaviorssuchascheating,andarecompetitiveincomparisonwithstandardmethods.Evenifthesuggestedmodelsdonotshowconsider-ableimprovement
8、soverothercompetingmodels(suchasBrockhoff&Skovgaard'smodel[12]),they,however,alsopermittoestimateinterestingfeaturesovertheraters–suchastheirreac
