An Unsupervised Approach for Detecting Group Shilling Attacks in Recommender Systems Based on Topological Potential and Group Behaviour Features

To protect recommender systems against shilling attacks, a variety of detection methods have been proposed over the past decade. However, these focus mainly on individual features and rarely consider lockstep behaviours among attack users, which suffer from low precision in detecting group attacks. In this work, we propose three-stage method based strong members behaviour for First, construct weighted user relationship graph by combining direct indirect collusive degrees between users. Second, find all dense subgraphs to generate set suspicious groups introducing topological potential method. Finally, use clustering detect extracting features. Extensive experiments Netflix sampled Amazon review datasets show that approach is effective attacks systems, F1-measure two can reach 99 percent 76 percent, respectively.