Research On The Techniques Of Social Network Priavacy Preserving Against Graph Structural Attacks

With the Internet and information technology rapid development, people have paid more attention to social network gradually. Through the research of social network, people can understand social phenomenon, predict human behavior, which provides greatly convenient for social structure analysis. But along with the social network research giving people convenient at the same time, more and more social network data is released to the web, which may include some of the privacy information. Attackers can use their existing background knowledge to steal personal privacy, resulting in the leakage of user privacy information. So along with the use of social network analysis, protecting user privacy without leakage has become a more and more attention problem. However the existing research doesnot consider the case where attackers have background knowledge of both vertex degrees and weights.At present, there are two kind of social network privacy protection methods. One is the clustering-based social network privacy protection method, and the other is the method of modifying social network privacy protection. They have the following assumptions on attackers’background knowledge:identifying attributes of vertices, vertex degrees, link relationship, neighborhoods, and embedded subgraphs. Therefore, we assume that the attackers have the background knowledge of both vertex degrees and weights, and deeply research on social network privacy protection problem.First, we cluster similar nodes in social network according to the social network feature of "small world" phenomenon and power-law distribution phenomenon. This step could modify original graph with less anonymization. For the similar vertices judegment, we propose the vertex anonymization information loss measurement method to judge the similarity of two vertices. Users could define K as the minimal number of vertices in each cluster. We take the most K similar vertices to construct a cluster. Then, we anonymize clusters based on the degrees of nodes and edges’ weights information to satisfy K-anonymity requirements. Because there is some difference among vertics’degrees and edges’weights, we modify original graph information as less as possible in order to make information loss low. We propose a method of equal edges matching. By the means of building the matching relation among edges whose weights are equal, we modify the edge which has much matching as less as possible. So it can reduce anonymization information loss effectively. We add edges to anonymize vertex’s degree. For edges’ weights anonymization, we change the original edge value to a new value. Finally, our experimental results on real datasets showed that the proposed social network privacy protection method not only reduces anonymization information loss effectively, but also protect user’s privacy effectively.