Privacy Protection in Memory-Based Collaborative Filtering

Abstract. We discuss the issue of privacy protection in collaborative filtering, focusing on the commonly-used memory-based approach. We show that the two main steps in collaborative filtering, being the determination of similarities and the prediction of ratings, can be performed on encrypted profiles, thereby securing the users ’ private data. We list a number of variants of the similarity measures and prediction formulas described in literature, and show for each of them how they can be computed using encrypted data only. Although we consider collaborative filtering in this paper, the techniques of comparing profiles using encrypted data only is much wider applicable.