With the development of social networks, more and more data about users are released on social platforms such as Facebook, Enron, WeChat, in terms of social graphs. Without the efficient anonymization, the graph data publishing will cause serious privacy leakage of users, for example, malicious attackers might launch 1-neighborhood graph attack on targets, which assumes that 1-hop neighbors and the relations among them are known by attackers, thereby, targets can be re-identified in anonymous social graphs. To prevent such attack, we propose a Graph Partition based Privacy-preserving Scheme, named GPPS,i n social networks to realize social graph anonymization. The proposed GPPS preserves users’ identity privacy by k-anonymity which achieved by node clustering and graph modification. Specifically, in the similarity matrix calculation, we introduce the degree-based graph entropy to improve the accuracy of node clustering. Then, the graph modification is implemented to achieve the k-anonymity of users and meanwhile minimize the graph information loss. The experiment results illustrate that the proposed GPPS is effective and efficient both on synthetic and real data sets.

随着社交网络的发展，越来越多的用户数据以社交图谱的形式发布在Facebook、安然、微信等社交平台上。如果没有有效的匿名化，图数据的发布会造成用户的严重隐私泄露，例如，恶意攻击者可能会对目标发起 1-邻域图攻击，假设攻击者知道 1-hop 邻居及其之间的关系，从而，目标可以在匿名社交图中重新识别。为了防止这种攻击，我们在社交网络中提出了一种基于图分区的隐私保护方案，称为GPPS，以实现社交图匿名化。提议的 GPPS 通过k 来保护用户的身份隐私-通过节点聚类和图修改实现的匿名性。具体来说，在相似度矩阵计算中，我们引入了基于度的图熵来提高节点聚类的准确性。然后，实现图修改以实现用户的k-匿名性，同时最小化图信息丢失。实验结果表明，所提出的 GPPS 在合成数据集和真实数据集上均有效且高效。