Online Pricing for Mobile Crowdsourcing with Multi-Minded Users

Kai Han, Yuntian He, Haisheng Tan, Shaojie Tang, He Huang, Jun Luo

Abstract: Mobile crowdsourcing has been proposed as a promising approach for urban data collection, but it has also brought the critical problem of designing proper mechanisms to incentivize user participation. Most previous work on crowdsourcing incentivization has assumed that each user holds a single private cost for participation or behaves in a "win all or nothing" (a.k.a. "single-minded") manner. However, in some crowdsourcing applications such as Amazon's Mechanical Turk, the users are usually "multi-minded" in the sense that each of them holds heterogeneous private costs for different tasks and only performs a portion of her/his interested tasks according to the announced payments. To address this problem, we propose LIME, an onLine prIcing mechanism to incentivize Multi-minded usErs under the scenario where the users arrive sequentially in an arbitrary order. We show that the design of LIME involves solving a "dummy semi-bandits with multiple knapsacks and random costs" problem, which has not been investigated before, and we also prove that LIME achieves several desirable properties including computational efficiency, budget feasibility, truthfulness, individual rationality and low regret on the utility. Finally, the effectiveness of LIME as well as its superiorities over prior related work are demonstrated through extensive simulations.