Robust Collaborative Filtering Based on Multiple Clustering
Jianwei Zhang · 2019
Collaborative filtering is widely used at online vendors and review sites in order to recommend items based on the ratings of many users. However, there are several problems with this method, and one of them is the existence of attacks that intend to distort the predicted ratings of specific items. This paper proposes a collaborative filtering method that reduces the impact of attacks while maintaining or improving the prediction accuracy by applying clustering to the target data multiple times and predicting the ratings for unrated items inside each cluster. In addition, the usefulness of the method is investigated using an evaluation method that measures the error between actual user ratings and predicted ratings. Furthermore, the robustness against attacks is investigated by comparing the prediction errors before and after attacks.