Research Seminar by ZHAO Yongjun | Protecting User Privacy in Private Set Intersection: A Journey Toward Mitigating User Tracking

Private Set Intersection (PSI) allows two parties, each holding an input set, to compute functions of corresponding values for shared set members, while maintaining the confidentiality of both the intersecting and non-intersecting elements. It has been used in applications such as ad providers and advertisers sharing privileged user behavior data to measure ad effectiveness, cloud storage operators detecting child exploitation material on users' encrypted cloud data, etc. A USENIX22' study found that common PSIs that disclose the intersection size can be vulnerable to attacks that exploit this disclosure, thereby revealing over 1% of one party's users to the other party in practical Ads measurement. To mitigate the risk of privacy leakage, we developed the DPCA-PSI protocol, which incorporates a novel PSI approach and a two-party differentially private (DP) mechanism, providing a secure means of computing intersection-related statistics from private datasets while preserving DP protection. DPCA-PSI is open-source and available to researchers and practitioners.