Crowdsourcing Task Design Using Multi-agent Systems and an Enhanced Genetic Algorithm

Crowdsourcing is a business model that assigns tasks to multiple online workers who complete them via the Internet. However, the anonymity of these workers presents a significant challenge for requesters when ensuring task quality. To improve task quality, we aim to automatically design crowdsourcing tasks tailored to requesters’ metrics. Experiments are conducted on Amazon mechanical turk (AMT) to identify the behaviors of online workers, forming a multi-agent system (MAS) as a testbed for evaluating and optimizing task design using an enhanced genetic algorithm. We also show how the MAS can create tasks that meet specified quality metrics. Finally, we validate our task designer through AMT experiments, paving the way for a data-driven approach to task quality assurance in crowdsourcing.