Distance-Sensitive Task Allocation for Mobile Crowdsensing: A Reinforcement Learning-Based Mechanism

Mobile crowdsensing services (MCSs) have long been a popular research topic in service computing. An MCS provider recruits users to complete data collection tasks, with an incentive mechanism employed to achieve greater utility. However, most existing studies have assumed that data collected by different users in the same area have the same value, which is not consistent with reality and reduces the quality of service. For example, noise or air quality data collected from different locations within the same area may differ considerably. In this paper, a distance factor is innovatively integrated into the MCS task allocation problem to more accurately represent real situations. Specifically, we transform the distance-sensitive online MCS task allocation problem into an integer programming model with budget constraints and ordered submodular characteristics. Additionally, we design a reverse auction mechanism based on reinforcement learning (RL) to solve the winner determination and payment problems. Unlike existing machine learning-based mechanisms, the proposed algorithms can strictly guarantee the economic characteristics of the auction mechanism, such as its individual rationality, truthfulness, and budget feasibility, through weight replacement and equilibrium pricing strategies. In experimental evaluations, we compare our algorithm with the offline optimal (OPT) algorithm, the advanced monotonic discrete online TDMC algorithm and the FULL SP-RL algorithm as benchmarks. The results indicate that our approach can enable service providers to obtain higher values with lower payments.