《Truthful Incentives in Crowdsourcing Tasks using Regret Minimization Mechanisms 使用后悔最小化机制的众包任务中的真实激励》由会员分享，可在线阅读，更多相关《Truthful Incentives in Crowdsourcing Tasks using Regret Minimization Mechanisms 使用后悔最小化机制的众包任务中的真实激励（11页珍藏版）》请在金锄头文库上搜索。
1、Truthful Incentives in Crowdsourcing Tasks using Regret Minimization Mechanisms Adish Singla ETH Zurich Zurich Switzerland adish.singlainf.ethz.ch Andreas Krause ETH Zurich Zurich Switzerland krauseaethz.ch ABSTRACT What price should be off ered to a worker for a task in an online labor market?How c
2、an one enable workers to ex- press the amount they desire to receive for the task com- pletion? Designing optimal pricing policies and determin- ing the right monetary incentives is central to maximizing requesters utility and workers profi ts. Yet, current crowd- sourcing platforms only off er a li
3、mited capability to the re- quester in designing the pricing policies and often rules of thumb are used to price tasks. This limitation could result in ineffi cient use of the requesters budget or workers becoming disinterested in the task. In this paper, we address these questions and present mecha
4、nisms using the approach of regret minimization in online learning. We exploit a link between procurement auc- tions and multi-armed bandits to design mechanisms that are budget feasible, achieve near-optimal utility for the re- quester, are incentive compatible (truthful) for workers and make minim
5、al assumptions about the distribution of work- ers true costs. Our main contribution is a novel, no-regret posted price mechanism, BP-UCB, for budgeted procure- ment in stochastic online settings.We prove strong the- oretical guarantees about our mechanism, and extensively evaluate it in simulations
6、 as well as on real data from the Mechanical Turk platform. Compared to the state of the art, our approach leads to a 180% increase in utility. Categories and Subject Descriptors K.4.4 Computers and Society: Electronic Commerce - Payment schemes; H.2.8 Database Management: Data- base applications -
7、Data mining General Terms Algorithms, Economics, Experimentation, Human Factors, Theory Keywords Crowdsourcing, incentive compatible mechanisms, procure- ment auctions, posted prices, regret minimization, multi- armed bandits Copyright is held by the International World Wide Web Conference Committee
8、 (IW3C2). IW3C2 reserves the right to provide a hyperlink to the authors site if the Material is used in electronic media. WWW 2013, May 1317, 2013, Rio de Janeiro, Brazil. ACM 978-1-4503-2035-1/13/05. 1.INTRODUCTION The growth of the Internet has created numerous oppor- tunities for crowdsourcing t
9、asks to online “workers”. Spe- cialized marketplaces for crowdsourcing have emerged, in- cluding Amazons Mechanical Turk (henceforth MTurk 1) and Click Worker1, enabling“requesters”to post HITs (Hu- man Intelligence Tasks), which can then be carried out by pools of workers available online and suita
10、ble for the task. Some of the tasks that are posted on these platforms in- clude image annotation, rating the relevance of web pages for a query in search engines, translation of text or tran- scription of an audio recording. Similarly, in platforms like social networks, users can be compensated for
11、 participation in a viral marketing campaign. The requester generally has a limited budget for the task and needs to come up with a payment scheme for workers in order to maximize the util- ity derived from the task. For workers, the main goal is to maximize their individual profi t by deciding whic
12、h tasks to perform and at what price. Monetary incentives in crowdsourcing tasks. One of the central components of these platforms is to design the right incentive structure and pricing policies for workers that maximize the benefi ts of both requester and the work- ers. Overpricing the tasks would
13、result in ineffi cient use of the requesters budget, whereas underpricing could lead to task “starvation” because of unavailability of the workers willing to participate. In this light, how can one design op- timal pricing policies? How can workers communicate and negotiate the price with requesters
14、? How would the mar- ket behave if workers act strategically by misreporting their costs for their benefi t? These are some of the questions that naturally come to mind while studying incentive structures for these online markets, yet they are not well understood. Pricing models. Current crowdsourci
15、ng platforms off er limited capability to the requester in designing the pricing policies, mostly limiting them to a single fi xed price (“fi xed price model”). One way to set prices under such models is to estimate workers costs via a market analysis and then compute an optimal fi xed price which w
16、ould maximize the utility. However, there are many diffi culties in inferring this optimal fi xed price, including the high cost of market sur- veys, the dynamic and online nature of the labor markets, inexperience of the requester and challenges in soliciting true costs from workers because of their self-interest. An alter- nate approach is to use tools of online procurement auctions where workers can bid on the price they are willing to receive 1 1167 and the requesters mec
