Cross-Task Crowdsourcing跨任务众包

Cross-Task Crowdsourcing跨任务众包

ID:18432653

大小:356.88 KB

页数:9页

时间:2018-09-17

Cross-Task Crowdsourcing跨任务众包_第1页
Cross-Task Crowdsourcing跨任务众包_第2页
Cross-Task Crowdsourcing跨任务众包_第3页
Cross-Task Crowdsourcing跨任务众包_第4页
Cross-Task Crowdsourcing跨任务众包_第5页
资源描述:

《Cross-Task Crowdsourcing跨任务众包》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、Cross-TaskCrowdsourcingKaixiangMo1,ErhengZhong1andQiangYang1,21:HongKongUniversityofScienceandTechnology,HongKong;2:HuaweiNoah’sArkLab,ScienceandTechnologyPark,Shatin,HongKong{kxmo,ezhong,qyang}@cse.ust.hkABSTRACTKeywordsCrowdsourcingisaneffectivemethodforcollectinglabeledCrowdsourcing,

2、TransferLearningdataforvariousdataminingtasks.Itiscriticaltoensuretheveracityoftheproduceddatabecauseresponsescollected1.INTRODUCTIONfromdifferentusersmaybenoisyandunreliable.PreviousworkssolvethisveracityproblembyestimatingboththeProducinglarge-scaletraining,validationandtestsetsisuser

3、abilityandquestiondifficultybasedontheknowledgevitalformanymachinelearninganddataminingapplica-ineachtaskindividually.Inthiscase,eachsingletaskneedstions.Mostoftenthistaskhastobecarriedout“byhand”largeamountsofdatatoprovideaccurateestimations.How-andthusitisdelicate,expensive,andtedious.

4、Crowd-1ever,inpractice,budgetsprovidedbycustomersforagivensourcingsystemssuchasAmazonMechanicalTurk,re-23targettaskmaybelimited,andhenceeachquestioncanbeCAPTCHAandtheESPgamehavemadeiteasytodis-presentedtoonlyafewuserswhereeachusercananswertributesimplelabelingtaskstohundredsofusers(ref

5、erredonlyafewquestions.ThisdatasparsityproblemcancausetoasworkerinMechanicalTurk).Crowdsourcingiswidelypreviousapproachestoperformpoorlyduetotheoverfit-usedintaskssuchassentimentclassification[12],objecttingproblemonraredataandeventuallydamagethedatarecognition[7],ranking[13]andclusterin

6、g[6],etc.veracity.Fortunately,inreal-worldapplications,userscanAtypicalcrowdsourcingpipelinecanbedividedintothreeanswerquestionsfrommultiplehistoricaltasks.Forexam-mainsteps:1:Taskdesign,2:Datadistributionand3:An-ple,onecanannotateimagesaswellaslabelthesentimentsweraggregation.Intheans

7、weraggregationstep,oneneedsofagiventitle.Inthispaper,weemploytransferlearning,toaggregatethesenoisyandunreliableresponsesintoatruewhichborrowsknowledgefromauxiliaryhistoricaltaskstoanswer.Answeraggregationismostimportantstepbecauseimprovethedataveracityinagiventargettask.Themoti-itis

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。