Abnormal Profiles Detection Based on Time Series and Target Item Analysis for Recommender Systems

Collaborative filtering (CF) recommenders are vulnerable to shilling attacks designed to affect predictions because of financial reasons. Previous work related to robustness of recommender systems has focused on detecting profiles. Most approaches focus on profile classification but ignore the group attributes among shilling attack profiles. Attack profiles are injected in a short period in order to push or nuke a specific target item. In this paper, we propose a method for detecting suspicious ratings by constructing a time series. We reorganize all ratings on each item sorted by time series. Each time series is examined and suspected rating segments are checked. Then we use techniques we have studied in previous study to detect shilling attacks in these anomaly rating segments using statistical metrics and target item analysis. We show in experiments that our proposed method can be effective and less time consuming at detecting items under attacks in big datasets.