Preserving Social Relationship Privacy via the Exponential Mechanism of Personalized Differential Privacy

Abstract

Presently, the majority of social networking platforms tend to outsource the analysis of social relationship data to third-party companies. Existing methods, which generally aim to protect social relationships by erasing friendship links or introducing uniform noise into datasets, do not take into account the risk of inference attacks or the actual privacy needs of users. To address these concerns, we present a novel method, named exponential mechanism of personalized difference privacy (EPDP), for preserving the privacy of social relationships, based on the EPDP. We develop specific social relationship indices to group friendship links and divided these links into distinct privacy levels, each with a unique privacy budget. Then, we select representative elements from each group using sampling and the exponential mechanism to generalize the original datasets, ensuring compliance with personalized difference privacy principles. Metrics for privacy and utility assessment are devised to evaluate method performance. Experimental results reveal that EPDP offers superior utility compared to uniform differential privacy (UDP) and provides better privacy protection than the state-of-the-art. Moreover, we explore the impact of various parameters on data utility. This article marks the pioneering effort to introduce a privacy-preserving method based on the exponential mechanism for the safeguarding of social relationships.