PPO-TA: Adaptive task allocation via Proximal Policy Optimization for spatio-temporal crowdsourcing

Abstract

With the pervasiveness of dynamic task allocation in sharing economy applications, the online bipartite graph matching has attracted people’s increasing attention to its research in recent years. Among its application in sharing economy, the crowdsourcing allocate the tasks to workers dynamically. There are still three main problems that need to be addressed from previous studies. (1) These task allocation algorithms usually ignore the long-term utility on crowdsourcing platforms. (2)The current research works show that it has low allocation numbers. (3) Due to the low number of allocations, it becomes difficult to improve total allocation utilities. In this paper, we considered the long-term utility and drawed an idea of dynamic delay bipartite graph matching(DDBM). We proposed a Policy Gradient Based Discrete Threshold Task Allocation algorithm (DTTA) and a Proximal Policy Optimization Based Continuous Threshold Task Allocation algorithm (PPOTA) to solve these problems. The extensive experimental results on two real datasets demonstrate that the proposed algorithms are superior to the existing algorithms in both effectiveness and efficiency.