Private and Efficient Set Intersection Protocol for Big Data Analytics

Abstract

Private Set Intersection (PSI) is a fundamental multi-party computation primitive used to secure many political, commercial, and social applications. PSI allows mistrustful parties to compute the intersection of their private sets without leaking additional information. PSI protocols has been largely proposed for both the semi-honest and the malicious settings. Nevertheless, the semi-honest setting does not suffice in many realistic scenarios and security in the malicious setting is built upon cryptographic schemes, which require hard assumptions and induce a high computational cost. In this work, we propose a novel twoparty PSI protocol secure under the mixed model, where the server may be semi-honest and the client may be malicious. We build our protocol upon matrix algebra without using any cryptographic schemes or nonstandard assumptions and we provide simulation-based security proof. Our protocol achieves a linear asymptotic complexity O(kv + kc) for communications and server computations, where kv and kc are sizes of participating parties' sets. Besides, we compare empirical performance of our solution to the insecure hashing solution used in practice. Experimental results reveal efficiency and scalability of our new PSI protocol, which makes it adequate for Big Data analytics.