De-Anonymizing Dynamic Social Networks

Online social network data are increasingly made publicly available to third parties. Recent studies show that it is possible to recover sensitive information from the released data and several anonymization techniques have been proposed to protect individual privacy. However, most of the existing defenses have focused on ``one-time´´ releases and do not take into consideration the re- publication of dynamic social network data. Re- publishing data periodically is a natural result of social network evolution and an emerging requirement of dynamic social network analysis. In this paper, we show that by utilizing correlations between sequential releases, the adversary can achieve high precision in de-anonymization of the released data, suppressing the uncertainty of re-identifying each release separately and synthesizing the results afterwards. Besides, we combine structural knowledge with node attributes to compromise graph modification based defenses. With experiments on real data, this work is the first to demonstrate feasibility of de-anonymizing dynamic social networks and should arouse concern for future works on privacy preservation in social network data publishing.