Spotting anomalous ratings for rating systems by analyzing target users and items

Abstract

Online rating systems play an important role in recommender systems. Collaborative filtering recommender systems are highly vulnerable to “shilling” attacks in reality. Although attack detection based on the attacks have been extensively researched over the last decade, the studies on this issue have not reached an end. Furthermore, only using the existing features is not easy to improve their detection performance. In this paper, we present an unsupervised detection method to defend such attacks, which consists of two stages. Based on the existing features of user and item, more effective features are selected using adaptive structure learning which takes advantage of adaptive local and global structure learning. In the first stage, suspected users are determined by exploiting a density-based clustering method based on the selected features. Then, the selected features of item are applied to find out suspicious items in order to further spot the concerned attackers based on the result of the first stage. Finally, the attackers can be detected. Extensive experiments on the MovieLens-100K dataset demonstrate the effectiveness of the proposed approach as compare to competing methods. It is noteworthy that discovering interesting findings including anomalous ratings and items on Amazon dataset also is investigated.