Abstract: The majority of real-world complex systems can be abstracted as dynamic information networks, such as social network, co-author network and e-mail network. Dynamic information networks are temporal, complex and changeable. To understand the behavioral trend of the complex systems, it is necessary to analyze the evolution of the network structures, especially the abnormal phenomena in the evolution. Aiming at detecting the anomalies in the dynamic evolution of the network structure, this paper utilizes “roles” to capture the structural characteristics of nodes, proposes the notion of role evolving outliers (REOutliers), and proposes a pattern-based role evolving outliers detection (P-REOD) method. This method mines the frequent patterns that roles of dynamic network structure evolve over time, and evaluates the degree of a node deviating from the frequent patterns to find the REOutliers. The experimental results show that the proposed method is highly effective in discovering interesting REOutliers.

Key words: dynamic networks, pattern mining, outlier detection

摘要： 现实世界中的社交网络、合作者网络、邮件网络等诸多复杂系统均可抽象为动态信息网络。动态信息网络具有时序、复杂、多变的特征，分析其网络结构随时间演化的过程，尤其演化过程中出现的异常现象，对理解复杂系统的行为倾向于演化趋势具有重要意义。致力于动态信息网络中异常结构演化过程的发现， 通过角色定义刻画网络的结构特征，提出了角色演化异常（role evolving outliers，REOutliers）的概念，并给出了基于模式挖掘的角色演化异常发现算法（pattern-based role evolving outliers detection，P-REOD）。该算法挖掘整个网络中角色随时间演化的频繁模式，通过比较节点到频繁模式的相异程度进行REOutliers发现。实验表明，该算法能够进行有效的角色演化异常发现。

关键词: 动态网络, 模式挖掘, 异常发现