A capsule-based reinforcement learning framework for supply-demand matching in mobile crowdsourcing

Mobile crowdsourcing services such as taxi and food delivery are playing an increasingly indispensable role in people's daily lives. Due to the complexity and dynamism in mobile scenarios, it is difficult to predict the trajectories and behavioral patterns of service providers and requesters indifferent regions at different times for effective scheduling in crowdsourcing platforms. The imbalance between supply and demand in different sub-regions has affected the sustainable operation of crowdsourcing services. To address this issue, we propose a capsule-based actor-critic reinforcement learning framework to achieve balanced supply and demand of mobile crowdsourcing services across multiple sub-regions. The framework includes a capsule network- based actor-critic reinforcement learning model and a willingness prediction model, which can continuously adjust the distribution between the supply and demand of mobile crowdsourcing services while taking into account the participants' daily trajectory preferences, thus improving the experience and satisfaction of mobile crowdsourcing participants and facilitating the sustainable operation of the platform. The proposed framework has been validated on several real-world mobile crowdsourcing datasets for its effectiveness.