User profile attack anomaly detection algorithm based on time series analysis

Collaborative filtering recommender systems are vulnerable to manipulation by malicious attacks, which can significantly affect the robustness of recommender systems. Aims at the defects and deficiencies of the previous attack detection methods, we propose an anomaly detection algorithm based on analyzing rating distribution characteristics of item over rating time series. First, we generate the rating time series by sorting the ratings based on the time stamps of each item; then partition the time series into several consecutive groups according to certain time interval, and calculate the sample average confidence interval of ratings for the item-self. We detect the item whether is under attack by monitor the rating behavior during the new coming time period. The experimental results show the effectiveness of our proposed algorithm. © 2010 Binary Information Press.