De-anonymization of Social Networks: the Power of Collectiveness

The interaction among users in different social networks raises deep concern on user privacy, as it may facilitate the assailants to identify user identities by matching the anonymized networks with a correlated sanitized one. Prior arts regarding such de-anonymization problem can be primarily divided into a seeded case or a seedless one, depending on whether or not there are a subset of pre-identified nodes. The seedless case is much more complicated since the adjacency matrix representation of one-hop user relations delivers limited structural information. To address this issue, we, for the first time, integrate the multi-hop neighborhood relationships, which exhibit more structural commonness between the anonymized and the sanitized networks, into seedless de-anonymization process. Our aim is to sufficiently leverage these multi-hop neighbors of all nodes and minimize the total disagreements of these multi-hop adjacency matrices, which we call collective adjacency disagreements (CADs), between two networks of different sizes. Theoretically, we demonstrate that CAD enlarges the difference between wrongly matched node pairs and correctly matched pairs, whereby two networks can be correctly matched with high probability even when the network density is below logn. Algorithmically, we adopt the conditional gradient descending method on a collective-form objective, which can efficiently find the minimal CADs for networks with broad degree distributions. Experiments on both synthetic and realworld networks return desirable de-anonymization accuracies thanks to the rich structural information manifested by such collectiveness, since most nodes can be correctly matched with their correspondences, especially in sparse networks where merely utilizing adjacency relations might fail to work.