UD-HMM: An unsupervised method for shilling attack detection based on hidden Markov model and hierarchical clustering

The existing unsupervised methods usually require a prior knowledge to ensure the performance when detecting shilling attacks in collaborative filtering recommender systems. To address this limitation, in this paper we propose an unsupervised method to detect shilling attacks based on hidden Markov model and hierarchical clustering. We first use hidden Markov model to model user's history rating behaviors and calculate each user's suspicious degree by analyzing the user's preference sequence and the difference between genuine and attack users in rating behaviors. Then we use the hierarchical clustering method to group users according to user's suspicious degree and obtain the set of attack users. The experimental results on the MovieLens 1 M and Netflix datasets show that the proposed method outperforms the baseline methods in detection performance.