Spammer Groups Detection Method Based on Local Expansion and Graph Contrastive Learning

The detection of spammer groups can help restrain the spamming behaviors, which is of great significance to ensure the credibility of review information on e-commerce platforms. However, the existing community detection-based methods for detecting spammer groups ignore the overlapping between spammer groups. Moreover, handcrafted indicators-based methods lack universality. Aiming at these problems, we propose a spammer groups detection approach based on local expansion and graph contrastive learning. First, we model the review dataset as a weighted user relationship graph by leveraging the user-product rating information and user-product-user meta- path information. Second, we design a local expansion-based overlapping community detection method to find candidate groups. Third, we apply graph contrastive learning to obtain the group vector representations and use local-global mutual information maximization to update model parameters. Finally, we utilize the isolation forest algorithm to calculate group suspiciousness and obtain spammer groups. The experimental results on Amazon and Yelp NYC datasets show that the proposed method has better detection performance than the baselines.