Using only the connection patterns gathered during a one-month period (comprising a stable population of 10,000 BitTorrent users), we extract communities of users that share interest in the same content. Despite the fact that connections in BitTorrent require not only shared interest in content, but also concurrent sessions, we find that strong communities of users naturally form – our analysis reveals that users inside the typical community are 5 to 25 times more likely to connect to each other than with users outside. These strong communities enable a guilt-by-association attack, where an entire community of users can be classified by monitoring one of its members. Our study shows that through a single observation point, an attacker trying to identify such communities can uncover 50% of the network within a distance of two hops.

To address this issue, we propose a new privacy-preserving layer for P2P systems that disrupts community identification by obfuscating users’ network behavior.