Efficient Task Decomposition in Crowdsourcing

Abstract

In order to facilitate crowdsourcing-based task solving, complex tasks are decomposed into smaller subtasks that can be executed either sequentially or in parallel by workers. These two task decompositions attract a plenty of empirical explorations in crowdsourcing. However the absence of formal study makes difficulty in providing task requesters with explicit guidelines on task decomposition. In this paper, we formally present and analyze those two task decompositions as vertical and horizontal task decomposition models. Our focus is on addressing the efficiency (i.e., the quality of the task’s solution) of task decomposition when the self-interested workers are paid in two different ways — equally paid and paid based on their contributions. By combining the theoretical analyses on worker’s behavior and simulation-based exploration on the efficiency of task decomposition, our study 1) shows the superiority of vertical task decomposition over horizontal task decomposition in improving the quality of the task’s solution; 2) gives explicit instructions on strategies for optimal vertical task decomposition under both revenue sharing schemes to maximize the quality of the task’s solution.