Exploiting Dynamic Privacy in Socially Regularized Recommenders

In this paper we introduce a privacy-aware collaborative filtering recommender framework which aims to address the privacy concern of profile owners in the context of social trust sparsity. While sparsity in social trust is mitigated by similarity driven trust using a probabilistic matrix factorization technique, the privacy issue is addressed by employing a dynamic privacy inference model. The privacy inference model exploits the underlying inter-entity trust information to obtain a personalized privacy view for each individual in the social network. We evaluate the proposed framework by employing an off-the-shelf collaborative filtering recommender method to make predictions using this personalized view. Experimental results show that our method offers better performance than similar non-privacy aware approaches, while at the same time meeting user privacy concerns.