WardpHMM: A Shilling Attack Detection Technique Using Wardp Method and Hidden Markov Model

Collaborative Filtering Recommender Systems (CFRSs) are widely employed in several applications because of its satisfying performance in the customized recommendation. Recent studies show that CFRSs are at risk of shilling attacks where attackers inject shilling profiles into the system. Malicious user injected ratings not only severely impact genuineness of recommendations but also user's trustworthiness within recommendation systems. Existing unsupervised clustering technique uses Ward method, which is an iterative method of low scalability. For addressing this issue, in this work an unsupervised SA detection technique named Ward(p)HMM has been proposed. It uses Ward(p) and Hidden Markov Model (HMM). In this proposed method HMM is used to measure difference of user's rating behavior. It generates User Suspicious Degree (USD) of each user by analyzing user's Suspicious Degree Range of Items (SDRI) and User's Matching Degree (UMD). Then Ward(p) method is applied to merge users based on USD and to acquire group of Attack Users (AUs). For performance analysis of the proposed technique, Amazon-ratings sample dataset was used. The performance comparison shows that proposed Ward(p)HMM technique outperforms baseline technique with respect to precision, recall and F1-score.