Preserving Social Network Privacy Using Edge Vector Perturbation

With the social network application, Popularity, the researchers can benefit through social network analysis, but it raises serious privacy concerns for the individual involved in social network. Some techniques have been proposed for protecting personal privacy. However, the existing methods tend to focus on un-weighted social network for anonymizing nodes and structure information or weighted social networks for anonymizing edge weight. We propose an edge vector perturbation method to preserve structural properties and edge weights for weighted social networks. First, we construct edge vector or edge space of the original weighted social network. Second, we calculate the edge betweenness and assign weights to elements in edge vector. Third, we construct release candidate set by the weighted Euclidean distance. We leverage the notions of edge vector and edge space in weighted social network. Given a social network G^s, we adopt two methods to build original edge vector E_Vec (G^s), and then select from some edge vectors from ψ(K_n)as publication candidate set of E_Vec(G^s). To ensure the effectiveness of released dataset, we use Euclidean distance between the vectors as metrics of the similarity. We execute experiments on datasets to study publication utility and quality. Our method can be applied to a typical perturbation algorithm to achieve better preservation of the utility of its output.