Online Rating Protocol Using Endogenous and Incremental Learning Design for Mobile Crowdsensing

Although mobile crowdsensing (MCS) has become a new paradigm of collecting, analyzing, and exploiting massive amounts of sensory data, how to effectively incentivize users’ active participation and high quality data contribution in MCS has become a critical issue. Existing offline incentive mechanisms are inefficient to be applied to the online scenario, where a large number of heterogeneous users arrive and leave dynamically in a random manner, meanwhile the MCS platform has to make an irrevocable immediate decision on whether to accept or reject users’ requests, and how much a user should be paid under a strict budget constraint without knowing future information. Although there are a few online incentive mechanisms, these efforts tend to focus on achieving some desirable properties without addressing the issues of online incentive as a whole, and are not effective to motivate users’ active participation and high-quality sensory data contribution simultaneously. In this paper, we integrate the quality of sensing, rating update, user selection, and payment determination to develop the first online rating protocol for practical MCS to deal with adverse selection and moral hazard simultaneously. First, we design an endogenous learning approach to quantify the users’ ratings based on their historical contributions. Second, we exploit an incremental learning approach to compute the density threshold based on users’ historical information. Third, we theoretically prove that our online rating protocol achieves the desirable properties of
computationally efficiency
,
budgetary feasibility
,
individual rationality
,
truthfulness
,
constant competitiveness
,
rating coverage
, and
incentive effectiveness
. Finally, extensive evaluation results support the performance of our protocol and validate the theoretical properties derived in this paper.