Allowing privacy-preserving analysis of social network likes

Social network Likes, as the 'Like Button' records of Facebook, can be used to automatically and accurately predict highly sensitive personal attributes. Even though this could be done for non malicious reasons, for example to improve products, services, and targeting, it represents a dangerous invasion of privacy with sometimes intolerable consequences. Anyway, completely defusing the information power of Likes appears improper. In this paper, we propose a mechanism able to keep Likes unlinkable to the identity of their authors, but to allow the user to choose every time she expresses a Like, those non-identifying (even sensitive) attributes she wants to reveal. This way, anonymous analysis relating Likes to various characteristics of the population is preserved, with no risk for users' privacy. The protocol is shown to be secure and also ready to the possible future evolution of social networks towards P2P fully distributed models. © 2013 IEEE.