Enabling Dynamic Analysis of Anonymized Social Network Data

Abstract

Anonymization is a widely used technique for the private publication of social network data. However, since the existing social network anonymization methods consider only one-time releases, they only reserve the static utility of the anonymized data. As social network evolves, these methods have posed severe challenges to the emerging requirement of dynamic social network analysis, which requires the dynamic utility of an evolving social network to be reserved for analysis. Instead of proposing a new anonymization method to handle dynamics, in this paper, we address these challenges by rebuilding connections between the sequentially published, anonymized data. By doing so, we have enabled a broad range of dynamic analysis to be applied to those already anonymized data without re-generating them. This suggests that our method is transparent to both the existing anonymization methods and the anonymized data. We adopt a combination of data-mining and graph-matching techniques to accomplish this task. The effectiveness of our method has been demonstrated on a series of real, dynamic social network data.