Research On Privacy Preserving Data Publishing In Social Network

In recent years, with the rapid development of network technology, the Internet haspenetrated into every aspect of human society, and impacts on people’s work, studyand life. The number of registered users of social network sites is rapidly increasingby the rise of Web2.0, resulting in an increasing number of large-scale social networkdata. In order to study and research the value of the data, a lot of social networkanalysis methods have been developed. However, if we can not properly use the socialnetwork data, publishing and studying the data may be a threat to privacy andinformation security.Many privacy principles and other models have been proposed to protect theprivacy of relational data. However, these principles can not be directly applied insocial network data. As compared with relational data, the social network data hassome links between different nodes. If we only anonymize the nodes, attackers canuse the prior information about existing structures to re-identify the label of each nodeor discover some private information.In this paper, we consider degree attack, subgraph attack, and edge-weightedattack which exists in social network data privacy protection, and then present somepractical solution.For degree attack, we use clustering algorithm to get k-anonymous vector, thendesign a greedy algorithm to construct k-anonymous social network which is based onHavel theorem. We use structural information loss to measure the data utility andsimulation experimental results show that anonymous social network has very highsimilarity to original graph.For subgraph attack, we propose a (d,k)-anonymization social network model,which uses iterative hash of the graph isomorphism detection methods and greedygraph anonymous matching strategy. We conducted many aggregate network querieson anonymous social network. The experimental results show that utility of theanonymous social networks is acceptable.For edge-weighted attack, we propose an edge-weighted perturbation model which bases on Dijkstra algorithm. It calls Dijkstra algorithm twice, and modifies theedge weight of social network with a small range, while keeping both the shortestpaths and the shortest distances between key node to target nodes remain unchanged.