資源描述：
《Besting the Quiz MasterCrowdsourcing Incremental Classification Games完勝問答大師：眾包遞增分類游戲》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、BestingtheQuizMaster:CrowdsourcingIncrementalClassi?cationGamesJordanBoyd-GraberBriannaSatinoff,HeHe,andHalDaumeIII`iSchoolandUMIACSDepartmentofComputerScienceUniversityofMarylandUniversityofMarylandjbg@umiacs.umd.edufbsonrisa,hhe,halg@cs.umd.eduAbstractbecauseprocessingtimeislimitedbyth
2、esheerquan-tityofdata,asinsiftinge-mailforspam(PujaraetCost-sensitiveclassi?cation,wherethefeaturesal.,2011).Insuchsettings,oftenthebestsolutionusedinmachinelearningtaskshaveacost,hasisincremental:allowadecisiontobemadewithoutbeenexploredasameansofbalancingknowl-seeingallofaninstance’sfe
3、atures.Wediscusstheedgeagainsttheexpenseofincrementallyob-incrementalclassi?cationframeworkinSection2.tainingnewfeatures.WeintroduceasettingOurunderstandingofhowhumansconductincre-wherehumansengageinclassi?cationwithmentalclassi?cationislimited.Thisisbecausecom-incrementallyrevealedfeatu
4、res:thecollegiatetriviacircuit.Byprovidingthecommunitywithplicatinganalreadydif?cultannotationtaskisoftenaweb-basedsystemtopractice,wecollectedanunwisetradeoff.Instead,weadaptarealworldtensofthousandsofimplicitword-by-wordsettingwherehumansarealreadyengaging(eagerly)ratingsofhowusefulfea
5、turesareforelicitinginincrementalclassi?cation—triviagames—andde-correctanswers.Observinghumans’classi?-velopacheap,easymethodforcapturinghumancationprocess,weimprovetheperformanceincrementalclassi?cationjudgments.ofastate-of-theartclassi?er.WealsousetheAfterqualitativelyexamininghowhuma
6、nscon-datasettoevaluateasystemtocompeteintheincrementalclassi?cationtaskthroughareduc-ductincrementalclassi?cation(Section3),weshowtionofreinforcementlearningtoclassi?cation.thatknowledgeofahuman’sincrementalclassi?-Oursystemlearnswhentoansweraquestion,cationprocessimprovesstate-of-the-a
7、rtrapaciousperformingbetterthanbaselinesandmosthu-classi?cation(Section4).Havingestablishedthatmanplayers.thesedatacontainaninterestingsignal,webuildBayesianmodelsthat,whenembeddedinaMarkovdecisionprocess,canengageineffectiveincremental1Introductionclassi?cation(Section5)
