Efficient Edge Anonymization of Large Social Graphs

Edges in a social graph may represent private information that needs to be protected. Due to their graph partition schemes, existing edge anonymization methods have several drawbacks, such as high utility loss and high computational overhead. In this paper, we present a new edge anonymization method, which partitions a graph using a new vertex equivalence relation called the neighbor-set equivalence. We show that whenever the resulting graph partition satisfies a graph confidence requirement, as the privacy measure, so does the graph partition based on automorphism. Our algorithm maintains the graph partition explicitly as a partition map and performs edge addition/removal in groups as merges of vertex classes on the partition map. To improve performance, our algorithm constructs a plan heuristically for a sequence of vertex class merges and executes the plan using a number of strategies. It checks for graph confidence only after the execution of the entire merge plan. We evaluate our method through extensive experiments using two real-world large social graphs and several graph utility measures. Our results show that the techniques used by our method effectively improve graph utility and reduce computational overhead.