《Task Matching in Crowdsourcing》由会员分享，可在线阅读，更多相关《Task Matching in Crowdsourcing（4页珍藏版）》请在金锄头文库上搜索。
1、Task Matchingin CrowdsourcingMan-ChingYuen1, Irwin King1,2, and Kwong-SakLeung11Departmentof ComputerScience and Engineering,The Chinese University of Hong Kong, Hong Kong2AT&T Labs Research,San Francisco,USAmcyuen, king, ksleungcse.cuhk.edu.hk;AbstractCrowdsourcing is evolvingas a distributed probl
2、em-solving and business production model in recent years. Incrowdsourcing paradigm, tasks are distributed to networkedpeople to complete such that a companys production cost canbe greatly reduced. A crowdsourcing process involves operationsof both requesters and workers. A requester submits a taskre
3、quest;a worker selectsand completesa task; and the requesteronly pays the worker for the successful completion of the task.Obviously, it is not efficient that the amount of time spent onselecting a task is comparablewith that spent on working on atask, but the monetaryreward of a task is just a smal
4、l amount.Literature mainly focused on exploring what type of tasks canbe deployed to the crowd and analyzing the performance ofcrowdsourcing platforms. However, no existingwork investigateson how to support workers to select tasks on crowdsourcingplatforms easily and effectively. In this paper, we p
5、ropose a novelidea on task matching in crowdsourcing to motivate workersto keep on working on crowdsourcing platforms in long run.The idea utilizes the past task preference and performance ofa worker to produce a list of available tasks in the order ofbest matching with the worker during his task se
6、lection stage.It aims to increase the efficiencyof task completion.We presentsome preliminaryexperimentalresultsin case studies.Finally, weaddress the possiblechallengesand discuss the future directions.Index Termscrowdsourcing; task model; task matchingalgorithmI. INTRODUCTIONNowadays, many tasks t
7、hat are trivial for humans continueto challengeeven the most sophisticatedcomputerprograms,such as image annotation. Therefore, these tasks cannot becomputerizedand they are traditionallyperformedby an em-ployee in a company. Crowdsourcingis a distributed problem-solvingand businessproductionmodel.I
8、n an articlefor Wiredmagazinein 2006, Jeff Howe defined“crowdsourcing”as “anidea of outsourcinga task that is traditionallyperformed byan employee to a large group of people in the form of anopen call” 2. The objective of crowdsourcing is to reducea companys productioncosts and make more efficient u
9、se oflabor and resources 3. An example of crowdsourcingtasksis the creative drawings, such as the Sheep Market 4. TheSheep Market is a web-basedartwork to implicatethousandsof workers in the creationof a massive databaseof drawings.Workers create their version of “a sheep facing to the left”using si
10、mple drawing tools. Each worker is responsiblefor adrawing receives a paymentof two cents for his labor.Because of the popularity of Web 2.0 technology, crowd-sourcing websites attract much attentions at present 10,9. Some popular examples of crowdsourcing websitesare Amazon Mechanical Turk (or MTur
11、k)1, CrowdFlower2,Taskcn3and TopCoder4. A crowdsourcingsite has two groupsof users: requestersand workers. The crowdsourcingsite ex-hibitsa list of availabletasks,associatingwithrewardandtimeperiod,that are presentedby requesters;and duringthe period,workers competeto provide the best submission.Mea
12、nwhile,a worker selects a task from the task list and completes thetask because the worker wants to earn the associatedreward.At the end of the period,a subsetof submissionsare selected,and the correspondingworkers are granted the reward by therequesters. In addition to monetary reward, a worker gai
13、nscredibilitywhen his task is acceptedby the requester.Nowadays, recommendationsystems11, 5, 8 are usedto suggestrelevant items(news, books,movies,etc.) attractingparticularusers on the Web. In these systems,assigningeachuser-item pair a score indicatingthe users rating on the item,based on which a
14、ranking list of items is generated to theuser as suggestions. In this paper, we address the issue ofrecommendationon crowdsourcingplatforms.Different fromtraditionalrecommendationsystems,multipleinformationandvarious relational dependencies in crowdsourcing systemsshould be utilized to improve recom
15、mendationresults.In this paper, we propose a novel task matching idea oncrowdsourcingplatforms. It can elicit the preference of taskworkers and collect their performancehistories. In this way,a list of tasks sorted by the order of best matching can beprovided to task workers when selecting tasks. It
16、 aims toincrease the efficiency of task completion.Task matching incrowdsourcingis importantbecauseof the following reasons: Motivate workers of diverse background to workon crowdsourcing tasks in long run. Currently, oncrowdsourcingsites,mostworkers onlyprovidemoderatecontributions 7 and there is a significantpopulationofyoungandwell-educatedIndianworkers 6.It canattractmore workers to contribute their efforts in long run if aworker find a suit
