Enhancing Game Policy Optimization in Mobile Crowdsourcing: A Reinforcement Learning Approach

Mobile Crowd Sensing (MCS) is a widely adopted approach for data collection across diverse applications. However, as the number of tasks and participants in MCS continues to grow, task allocation and dynamic pricing challenges have become increasingly complex. Existing research primarily focuses on single-task allocation problems, often overlooking the diversity and complexity inherent in multi-task, multi-worker scenarios. To address these challenges, this paper proposes the Enhanced Heuristic Search with Tabu and Local Search (EH-STLS) algorithm, alongside a Kolmogorov-Arnold Deep Q Network (KDQN) model, both grounded in a Stackelberg game framework. The EH-STLS algorithm employs a multi-objective optimization framework that combines tabu search and local search strategies to improve the efficiency of worker-task matching while ensuring high task quality. The KDQN model views task publishers as leaders and treats crowdsourcing platforms, encryption agencies, and workers as followers to achieve optimal dynamic pricing and utility allocation. Extensive experiments on synthetic datasets generated from real-world data reveal that the proposed methods substantially outperforms the baseline algorithms regarding task allocation quality and pricing efficiency, achieving up to a 21.7% increase in task quality and a 16.4% improvement in pricing effectiveness.