Using Genre Interest of Users to Detect Profile Injection Attacks in Movie Recommender Systems

While the popularity of recommender systems is growing rapidly in e-commerce services, profile injection attacks are a great threat to their robustness and trustworthiness. Such attacks can be easily produced and inserted in recommender systems to alter the recommendation results. In such systems, attackers intentionally insert attack profiles to change the systems output to their advantage. This paper presents the idea of utilizing a set of genre attributes in order to discriminate between attack and genuine profiles in a movie recommender system. Since attackers typically assign random ratings to the movies in attack profiles, the genre interest of attackers and genuine users who rate movies based on their preferences are different. Based on this idea, we build a system using genre attributes as inputs to a feed forward neural network in order to detect attackers. The performance of our proposed approach is presented and compared to other detection approaches. The results declare superiority of our proposed approach from precision and recall point of view.