Privacy-Preserving Similarity Computation in Cloud-Based Mobile Social Networks

Abstract

A growing number of mobile social network applications are taking advantage of cloud computing to store profiles of end users and run protocols which are compute-intensive. We focus on an application scenario of similarity computation between two users. To protect data security and privacy, mobile users encrypt their sensitive profiles before outsourcing to the cloud and different users choose different encryption keys. Three challenges need attention - how to compute on encrypted profiles under different keys, how to allow mobile users to stay offline during execution of the protocol, and how to select similarity metric. Existing schemes either rely on multi-key fully homomorphic encryption with one server or partially homomorphic encryption with two non-colluding servers. To balance computational complexity on one server and communication overhead between two servers, we put forward a privacy-preserving similarity computation protocol which supports both homomorphic additions and one homomorphic multiplication. We conduct security analysis and experimental evaluation of our scheme. The results show that our protocol is provably secure and runs reasonably fast, and thus can be applied in practice.