Abstract: Social network data publication is dynamic and unsafe. In order to avoid associated attacks using social network data at different time and take into account the diversity of node attributes, this paper proposes a method of privacy preserving in dynamic social network data publication. Firstly, nodes are clustered according to anonymous rules, and the anonymous graph for the time being is generated, in which the diversity of nodes in the same anonymous set is maximized, and leakage probability for both node attributes and edge relationships are less than 1/k. Secondly, the difference set of the data relation graph of the adjacent time is generated. Considering the anonymous graph of current time, the nodes and edges which do not exist in previous graph are deleted and published data are updated reversely. In consequence, the graph structures in different timestamps should be similar to improve the resilience to associated attacks. Finally, this paper evaluates the security and availability on the Sina Weibo data set and the public email correspondence data set. The experiment results show that the method meets users' personalized requirements of data privacy protection and data availability.

Key words: dynamic social network, privacy protection, anonymous rules, K-anonymity, data publication

摘要： 社会网络数据发布具有动态性与不安全性，为避免使用不同时刻的社会网络数据进行关联攻击，兼顾节点属性多样性，提出了一种动态社会网络数据发布隐私保护方法。首先，根据匿名规则进行节点聚类，求解当前时刻的匿名图，保证同一个匿名集中节点属性多样性最大的前提下，数据发布后的节点属性与边的泄露概率均小于1/k。然后，生成相邻时刻数据关系图的差集，结合当前时刻的匿名图，删除前序时刻不存在的节点与边，逆向更新已发布数据，保证不同时刻下的匿名图具有相似的图结构，抵御关联攻击。最后，采用新浪微博数据和邮件往来数据进行实验验证，对所提方法的安全性和可用性进行评估。实验结果表明所提方法兼顾了用户数据隐私保护和数据可用性的个性化需求。

关键词: 动态社会网络, 隐私保护, 匿名规则, K-匿名, 数据发布