I. Introduction
With the great popularization of mobile devices and wireless communications, Mobile CrowdSensing (MCS) [2] has rapidly become a powerful paradigm, which recruits a number of mobile users to cooperatively perform various urban sensing and computing tasks, e.g., traffic estimation, indoor-outdoor localization, and map semantics identification. In general, the MCS applications should provide proper rewards for the recruited users in order to cover their costs and encourage the participation [3]. However, considering the limited budget, we usually cannot afford all of the users, but have to recruit those who can complete the tasks more effectively, which raises the fundamental user recruitment problem in MCS [4], [5].