Trust-Driven Data Collection in Mobile Crowdsensing under Strategic User Behavior

Mobile crowdsensing (MCS) is an innovative sensing paradigm that leverages the widespread use of human-carried mobile devices, such as smartphones, to collect environmental data via built-in sensors and upload it to a central server. Compared to traditional static sensor-based approaches, MCS offers advantages such as broader coverage, greater flexibility, and lower deployment costs, driven by human mobility and the growing prevalence of mobile devices. A critical challenge in MCS lies in selecting participants based on task requirements and user states. Most prior research assumes that users honestly report their local states, such as their matching scores with task requirements when uploading data to the server. However, in real-world scenarios, strategic users may misrepresent this information to maximize their rewards. In this paper, we propose an MCS system where the server selects participants and determines their payoffs by progressively evaluating user reliability. Specifically, we model the strategic behavior of users using the Bayesian Persuasion framework. We formulate a joint optimization problem that integrates participant selection and data size requests allocation to maximize overall profit and sensing coverage, even in the presence of strategic user behavior. To address this problem, we develop a reinforcement learning solution based on Proximal Policy Optimization (PPO), enabling the server to dynamically select participants and adaptively update their reliability scores through Bayesian inference. Extensive simulations demonstrate that our PPO-based Bayesian approach significantly enhances data quality and sensing coverage compared to baseline methods.