Real-Time Privacy Risk Quantification in Online Social Networks

Matching the anonymous profile of an individual in an online social network (OSN) to their real identity raises serious privacy concerns as one can obtain sensitive information about that individual. Previous work has formulated the profile matching risk in several different ways and has shown that there exists a non-negligible risk of matching user profiles across OSNs. However, they are not practical to convey the risk to OSN users in real-time. In this work, using the output of such formulation, we model the profile characteristics of users that are vulnerable to profile matching via machine learning and make probabilistic inferences about how the vulnerabilities of users change as they share new content in OSNs (or as their graph connectivity changes). We evaluate the generated models in real data. Our results show that the generated models determine with high accuracy whether a user profile is vulnerable to profile matching risk by only analyzing their publicly available information in the anonymous OSN. In addition, we develop optimization-based countermeasures to preserve the user's privacy as they share their OSN profile with third parties. We believe that this work will be crucial for OSN users to understand their privacy risks due to their public sharings and be more conscious about their online privacy.