Preserving the Privacy of Social Recommendation with a Differentially Private Approach

With the popularity of social networks such as Facebook and twitter, social recommendations have become possible, which rely on individual's social connections in order to make personalized recommendations of ads, content, products, and people. Since recommendations involving sensitive information, adversaries may re-identify a user's sensitive information from the recommendation results using background information. This paper proposes a privacy preserving approach to address the problem in the context of social recommendation in a strict privacy notion, called differential privacy. The approach incorporates a clustering method to group users according to the structure of the target social network. Then use the weighted paths as the utility function, which measures the recommendation utility. It adds Laplace noise to the weight of social graph to inject perturbation. Experimental analysis are provided to show the proposed approach can ensure differential privacy while retaining the utility of social recommendation.