Composite Task Selection with Heterogeneous Crowdsourcing

A common feature among many crowdsourcing applications is to decompose the huge or complex tasks into some small sub-ones, which require some users with different skills to implement. The kind of tasks are composite and called Composite Tasks (CTs), which are said to be completed and return reward only after all of their sub-tasks are finished successfully. Meanwhile, users may have various capabilities to implement diverse sub-tasks (STs) with corresponding cost so users are heterogeneous. In this paper, we study the problem of how users choose the STs to maximize their payoff (reward minus cost) when there are multiple such CTs. This payoff maximization problem with multiple CTs and heterogenous users turns out to be NP-completed. We then propose a Local Composite Task Selection (LCTS) algorithm to help the users choose their subtask strategies. Its convergence and complexity are analyzed theoretically. For comparison, we design a centralized Composite Task Selection (CTS) algorithm and a Low Cost and Random sub-task selection (LCR) algorithm as benchmarks. Numerical results suggest that the LCTS algorithm achieves a similar payoff and task completion ratio to the CTS when the number of users is large. The performance of LCTS is highly over LCR on both of the payoff and the task completion ratio. The results also illustrate the quick convergence of the LCTS algorithm.