PPSI: Practical Private Set Intersection Over Large-Scale Datasets

Abstract

Private Set Intersection (PSI) is a fundamental building block in data analytics, which has extensive practical applications including Genome matching, Botnet detection, Social networking, etc. The continuous increase in large scale datasets makes traditional PSI protocols no longer scalable and efficient enough in practice. Thus, it becomes a promising problem to design an efficient private set intersection protocol over largescale datasets in a privacy-preserving manner. Unfortunately, some existing solutions in traditional two-party setting are not suitable for the weak computational capability clients due to multiple interactions and several solutions in server-aided (or outsourced) setting are not efficient enough for large-scale datsets (e.g., million or billion elements set size). In this paper, we construct two basic secure and practical PSI (PPSI) protocols over large-scale datasets in the server-aided setting based on Bloom filter. The first protocol is secure against a semi-honest server, while the second one is even secure against a malicious server who is able to arbitrarily add or remove some elements from the computed intersection without revealing intersection size to the server. Experimental results show that our two basic protocols only need around 15 seconds and 24 seconds (128-bit security in parallel mode) over one millionelement datasets respectively. Finally, we propose a multi-round mechanism for our two basic protocols, which as results show, can double the efficiency.