Differential Privacy for Context-Aware Recommender Systems

Abstract

How to prevent the individual privacy from being disclosed and incorporate contextual information into recommendations process is an urgent problem that needs to be solved in recommendation systems. Challenged by the above, a context-aware recommendation method that integrates Differential Privacy and Bayesian Network technologies is proposed. Firstly, in order to alleviate sparsity of the rating matrix, the paper adopts k-means algorithm to cluster items. And then add noises to ratings to protect users' privacy. Finally, the probability that a user likes a certain type of project in contextual information is calculated by Bayesian formula. Experimental evaluations show that the proposed algorithm can provide a stronger privacy protection while improving the accuracy of recommendations.