In the Web 2.0, more and more personal data are released by users (queries, social networks, geo-located data, ...), which create a huge pool of useful information to leverage in the context of search or recommendation for instance. In fully decentralized systems, tapping on the power of this information usually involves a clustering process that relies on an exchange of personal data (such as user proles) to compute the similarity between users. In this internship, we address the problem of computing similarity between users while preserving their privacy and without relying on a central entity, with regards to a passive adversary.