Paper 2016/948 Original: eprint.iacr.org/2016/948 ↗

Secure Computation in Online Social Networks

Abstract

Apart from their numerous other benefits, online social networks (OSNs) allow users to jointly compute on each other’s data (e.g., profiles, geo-locations, medical records, etc.). Privacy issues naturally arise in this setting due to the sensitive nature of the exchanged information. Ideally, nothing about a user’s data should be revealed to the OSN provider or "non-friend" users, and even her "friends" should only learn the output of a joint computation. In this work we propose the first security framework to capture these strong privacy guarantees for general-purpose computation. We also achieve two additional attractive properties: users do not need to be online while their friends compute on their data, and any user value uploaded at the server can be repeatedly used in multiple computations. We formalize our framework in the setting of secure multi-party computation (MPC) and provide two instantiations: the first is a non-trivial adaptation of garbled circuits that converts inputs under different keys to ones under the same key, and the second is based on two-party mixed protocols and involves a novel two-party re-encryption module. We experimentally validate the efficiency of our instantiations using state-of-the-art tools for two concrete use-cases.