Structural Diversity for Privacy in Publishing Social Networks

How to protect individual privacy in public data is always a concern. For social networks, the challenge is that, the structure of the social network graph can be utilized to infer the private and sensitive information of users. The existing anonymity schemes mostly focus on the anonymity of vertex identities, such that a malicious attacker cannot associate an user with a specific vertex. In real social networks, however, each vertex is usually associated with not only a vertex identity but also a community identity, which could represent the private information for the corresponding user, such as the political party affiliation or disease information sensitive to the public. In this paper, we first show that the attacker can still infer the community identity of an user even though the graph is protected by previous anonymity schemes. Afterward, we propose the structural diversity, which ensures the existences of at least k communities containing vertices with the same degree for every vertex in the graph, to provide the anonymity of the community identities. Specifically, we formulate a new problem, k-Structural Diversity Anonymization (k-SDA), which protects the community identity of each individual in publishing social networks. We propose an Integer Programming formulation to find the optimal solutions to k-SDA. Moreover, we devise three scalable heuristics to solve the large instances of k-SDA with different perspectives. The experiments on real data sets demonstrate the practical utility of our privacy model and our approaches.