Identity vs. attribute disclosure risks for users with multiple social profiles

Individuals sharing data on today's social computing systems face privacy losses due to information disclosure that go much beyond the data they directly share. Indeed, it was shown that it is possible to infer additional information about a user from data shared by other users-- this type of information disclosure is called attribute disclosure. Such studies, however, were limited to a single social computing system. In reality, users have identities across several social computing systems and reveal different aspects of their lives in each. This enlarges considerably the scope of information disclosure, but also complicates its analysis. Indeed, when considering multiple social computing systems, information disclosure can be of two types: attribute disclosure or identity disclosure-- which relates to the risk of pinpointing, for a given identity in a social computing system, the identity of the same individual in another social computing system. This raises the key question: how do these two privacy risks relate to each other? In this paper, we perform the first combined study of attribute and identity disclosure risks across multiple social computing systems. We first propose a framework to quantify these risks. Our empirical evaluation on a real-world dataset from Facebook and Twitter then shows that, in some regime, there is a tradeoff between the two information disclosure risks, that is, users with a lower identity disclosure risk suffer a higher attribute disclosure risk. We investigate in depth the different parameters that impact this tradeoff.