Improving Crowdsourcing Efficiency Based on Division Strategy

Abstract

This paper examines the efficiency in crowd sourcing, especially crowd sourcing for the software bug detection. Crowd sourcing has recently emerged as a lucrative paradigm for leveraging the collective intelligence of crowds. However, it has inherent weakness that a simple reward setting causes an uneven distribution of workers on each task, which reduces the efficiency of solving the tasks. A challenge is that the system designer is not allowed to set the reward to the arbitrary value because so-called "market wages" exist and if the reward is set to the value lower than the market wage, such a task fails to attract the sufficient number of workers. To solve this problem, we focus on the division strategy that divides the crowds into different groups and the workers compete with each other among the same group. We have developed a model that crowds write their codes independently and then try to find bugs in the codes written by the others. Next, we examine two division strategy, random grouping and ability grouping by analyzing the equilibrium behavior of each worker and carrying out simulations. The results show that the division strategy affect the efficiency of crowd sourcing bug detection and the random grouping leads to a higher efficiency compared to the ability grouping.