1Privacy-preserving Friendship-based Recommender Systems

Abstract—Today, recommender systems are playing an indispensable role in our daily life. However, nothing is for free – such systems have also upset the society with severe privacy concerns. In this paper, we first revisit the concept of computing recommendations based on inputs from both a user’s friends and a set of randomly chosen strangers. We propose two security models to formalize information leakages in recommender systems. We then clarify two protocols by Tang and Wang at ESORICS 2015, analyse their security in our security models, and investigate their performances according newly-constructed Twitter datasets and MovieLens 100k dataset. Our experiments show that the single prediction protocol is efficient and can be considered practical in reality. We finally propose a new decentralized single prediction protocol and compare it to the centralized (clarified) protocol. I