Privacy-Preserving Item-Based Collaborative Filtering Using Semi-Distributed Belief Propagation

Recommender systems are increasingly employed by e-commerce websites to suggest items to users that meet their preferences. Collaborative Filtering (CF), as the most popular recommendation algorithm, exploits the collected historic user ratings to predict ratings on unseen items for users. However, traditional recommender systems are run by the commercial websites, and thus users have to disclose their personal rating data to the websites in order to receive recommendations. This raises the privacy issue, as user ratings can be used to reveal sensitive personal information. In this paper, we propose a privacy-preserving item-based CF recommender system using semi-distributed Belief Propagation (BP), where rating data are stored at the user side. Firstly, we formulate the item similarity computation as a probabilistic inference problem on the factor graph, which can be efficiently solved by applying the BP algorithm. To avoid disclosing user ratings to the server or other user peers, we then introduce a semi-distributed architecture for the BP algorithm, where only probabilistic messages on item similarity are exchanged between the server and users. Finally, an active user locally generates rating predictions by averaging his own ratings on items weighted by their similarities to unseen items. As such, the proposed recommender system preserves user privacy without relying on any privacy techniques, e. g., obfuscation and cryptography. Further, there is no compromise in recommendation performance compared to the centralized counterpart of the proposed algorithm. Through experiments on the MovieLens dataset, we show that the proposed algorithm achieves superior accuracy.