Spammer detection via ranking aggregation of group behavior

Online review networks have become a popular platform for disseminating information which attracts large numbers of spammers to influence the reputation of the online products by writing fake reviews. Thus, the spammer detection becomes crucial in controlling the fraud reviews which is valuable for practical applications. However, existing ranking-based methods usually focus on the rating value and textual features of reviews ignoring the features of group behavior which results in the sharp degradation of spammer detection’s performance in case of large-scale spammer group attack. In this paper, we propose a novel ranking aggregation method based on features of collusive attack of spammer groups to optimize the spammer ranking algorithm by reassigning the spamicity score for each reviewer. Extensive experiments on various real-world datasets with spammer injection demonstrate that our proposed optimization method can significantly improve the accuracy and robustness of the spammer detection. The average improvements of performance for ranking before and after optimization achieve 10.73%, 42.86% and 4.32% respectively on the Recall, Precision and AUC metrics.