Effectiveness of Proximity-Based Outlier Analysis in Detecting Profile-Injection Attacks in E-Commerce Recommender Systems

Abstract

E-Commerce recommender systems are vulnerable to different types of profile-injection attacks where a number of fake user profiles are inserted into the system to influence the recommendations made to the users. In this paper, we have used two proximity-based outlier detection strategies in identifying fake user profiles inserted into the recommender system by the attacker. The first strategy that has been used in detecting attack profiles is a k-Nearest Neighbor based algorithm. The second strategy used is a clustering-based algorithm in generating outlier score of each user profile in the system database. Three attack models namely random attack, average attack and bandwagon attack model have been considered for our analysis. Performance of the k-Nearest Neighbor-based and clustering-based outlier detection strategies have been analyzed for different attack percentages and different filler percentages of the attack profiles.