Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记

Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记

ID:18440441

大小:270.00 KB

页数:6页

时间:2018-09-18

Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记_第1页
Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记_第2页
Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记_第3页
Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记_第4页
Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记_第5页
资源描述:

《Semi-Supervised Consensus Labeling for Crowdsourcing众包的半监督一致性标记》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、Semi-SupervisedConsensusLabelingforCrowdsourcingWeiTangMatthewLeaseDepartmentofComputerScienceSchoolofInformationTheUniversityofTexasatAustinTheUniversityofTexasatAustinwtang@cs.utexas.eduml@ischool.utexas.eduABSTRACTtasksthataredifficulttoeffectivelyautomatebutcanbeperformedbyremot

2、eworkers.Becauseindividualcrowdworkersoftenexhibithighvari-OnMTurk,“requesters”typicallysubmitmanyannota-anceinannotationaccuracy,weoftenaskmultiplecrowdtionmicro-tasks,andworkerschoosewhichtaskstoperform.workerstolabeleachexampletoinferasingleconsensusRequestersobtainlabelsmoreq

3、uicklyandaffordably,andlabel.Whilesimplemajorityvotecomputesconsensusbyworkersearnafewextrabucks.Unfortunately,accuracyofequallyweightingeachworker’svote,weightedvotingas-individualcrowdworkershasoftenexhibitedhighvariancesignsgreaterweighttomoreaccurateworkers,whereaccu-inpaststu

4、diesduetofactorslikepoordesignorincentivesofracyisestimatedbyinner-annotatoragreement(unsuper-tasks,ineffectiveorunengagedworkers,orannotationtaskvised)and/oragreementwithknownexpertlabels(super-complexity.Twocommonmethodsforqualitycontrolare:vised).Inthispaper,weinvestigatetheann

5、otationcostvs.(a)workerfiltering[6](i.e.identifyingpoorqualityworkersconsensusaccuracybenefitfromincreasingtheamountofandexcludingthem)and(b)aggregatinglabelsfrommulti-expertsupervision.Tomaximizebenefitfromsupervision,pleworkersforagivenexampleinordertoarriveatasingleweproposeasemi

6、-supervisedapproachwhichinfersconsen-“consensus”label.Inthispaper,wefocusontheconsensussuslabelsusingbothlabeledandunlabeledexamples.Weproblem;ourfutureworkwillstudyacombinedapproach.compareoursemi-supervisedapproachwithseveralexistingAccuratelyestimatingconsensuslabelsfromindivi

7、dualunsupervisedandsupervisedbaselines,evaluatingonbothworkerlabelsischallenging.AcommonapproachtothissyntheticdataandAmazonMechanicalTurkdata.ResultsproblemissimpleMajorityVoting(MV)[14,13,16],whichshow(a)averymodestamountofsupervisioncanprovideiseasytouseandcanoftenachieverelat

8、ivelygoodempir-significantbenefit,and(b)co

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。