Privacy-preserving Collaborative Filtering based on Randomized Perturbation Techniques and Secure Multiparty Computation

With the evolution of the Internet, collaborative filtering techniques are becoming increasingly popular in E-commerce recommender systems. Such techniques recommend items to users employing similar users' preference data. People use recommender systems to cope with information overload. Although collaborative filtering systems are widely used by E-commerce sites, they fail to protect users' privacy. Since many users might decide to give false information because of privacy concerns, collecting high quality data from users is not an easy task. Collaborative filtering systems using these data might produce inaccurate recommendations. To reserve privacy in collaborative filtering recommender systems, this paper presented a collaborative filtering algorithm based on randomized perturbation techniques and secure multiparty computation. The randomized perturbation techniques are used in the course of user data collection and can generate recommendations with decent accuracy. Employing secure multiparty computation to protect the privacy of collaborative filtering by distributing the user profiles between multiple repositories and exchange only a subset of the profile data, which is useful for the recommendation. Theoretical analysis shows that the algorithm based on randomized perturbation techniques and secure multiparty computation not only protect the users’ privacy, but also can keep the accuracy.