Inferring Attributes Using Centrality Measures

Information privacy has become one of the most urgent research issues in building next-generation information systems. There is a number of research efforts which is devoted to protect privacy. These efforts effectively prevent (or deny) only direct disclosure. While blocking direct disclosure enhances privacy, techniques based on statistical inference are also devised alongside to effectively work as indirect disclosure. An indirect disclosure is achieved by intelligently combining pieces of seemingly related or unrelated information. Specifically, in scenarios like social networks, where individuals have set some personal attributes as private, through their ties indirect disclosure can be achieved from common or shared attributes. A lot of work is already done regarding attributes inference in online social network. In this paper, we proposed techniques that utilize the centrality measures of graph for attributes inference. We have used two centrality measure "closeness" and "betweenness" for inferring missing attributes first, then used hybrid "Bayesian" and "local" techniques. Best of our knowledge no one have used centrality measure for attributes inference in online social network.