Enhancing social matrix factorization with privacy
Shahab Mokarizadeh, Nima Dokoohaki, Mihhail Matskin, Ramona Bunea · 2013
Within the course of this manuscript we present a privacy-preserving collaborative filtering recommender system which aims at alleviating the concern with privacy of user profiles within the context of sparse social trust data. While problem of sparsity in social trust is often addressed by taking similarity driven trust measures through a probabilistic matrix factorization technique, we address the issue of privacy by proposing a dynamic privacy inference model. The privacy inference model exploits the underlying inter-entity trust information in order to build a personalized privacy perspective for each individual within the social network. This is followed by our evaluation of the proposed solution by adopting an off-the-shelf collaborative filtering recommender library, in order to generate predictions using this personalized view.