A social network analysis of fraud prediction on crowdsourcing platforms

Abstract

In the context of crowdsourcing contests, where winners take all, attracting high-quality solvers and solutions presents a significant challenge. A key issue in this environment is protecting solvers’ intellectual property and preventing fraud risks such as solution plagiarism and theft. Addressing these challenges is essential for maintaining the integrity of the platform and encouraging innovation. This study applies social network analysis to examine the structural characteristics of fraudulent seekers and investigate whether they exhibit distinct social network features compared to legitimate users. Specifically, we focus on centrality, cohesion, and structural equivalence to identify potential markers of fraudulent intent. Using a dataset from 9,282 contest projects initiated in China in 2014, involving 6,241 active users and 246 fraudulent seekers, we tested a fraud detection model based on social network metrics. The results reveal significant differences in degree centrality, betweenness centrality, closeness centrality, and clustering coefficients between fraudulent and non-fraudulent nodes. The findings demonstrate that social network features, particularly centrality measures, can effectively differentiate fraudulent seekers from legitimate users. This study contributes to the theoretical understanding of fraud detection in crowdsourcing and offers practical insights for the development of more robust fraud detection strategies.