社群演化的稳健迁移估计及演化离群点检测

时序数据集中的社群演化模式是网络行为动力学研究与应用的重要领域.基于社群演化的离群点检测不仅能够发现新颖的异常行为模式,同时也有利于更准确地理解社群的演化趋势.运用成员关于社群隶属关系的变化,提出了社群演化迁移矩阵的概念,研究并揭示了迁移矩阵的若干性质及其与社群结构演化之间的关系.在采用稳健回归M-估计方法进一步优化迁移矩阵降低异常点干扰的同时,对社群演化离群点加以刻画和定义.鉴于复杂网络包含大量随机游走的边缘个体,所定义的离群点综合考虑其在社群中角色的变化和相对于社群总体迁移模式的差异.基于上述思想提出的演化离群点检测算法能够适应各类社群演化趋势,更有效地聚焦和发现大规模社会网络中重要成员的异常演化行为.实验结果表明,所提出的方法能够从大规模社会网络演化序列中发现重要的离群演化模式,并在现实中找到合理的解释.

Robust Transition Estimation for Community Evolution and Evolutionary Outlier Detection

Community evolutionary pattern analysis in temporal datasets is a key issue in social network dynamics research and applications. Identifying outlying objects against main community evolution trends is not only meaningful by itself for the purpose of finding novel evolution behaviors, but also helpful for better understanding the mainstream of community evolution. Upon giving the belonging matrix of community members, this study defines a type of transition matrix to characterize the pattern of the evolutionary dynamic between two consecutive belonging snapshots. A set of properties about the transition matrix is discussed, which reveals its close relation to the gradual community structural change in quantity. The transition matrix is further optimized using M-estimator Robust Regression methods by minimizing the disturbance incurred by the outliers, and the abnormality of the outlier objects can then be computed at the same time. Considering that large proportion of trivial but nomadic objects may exist in large datasets like those of complex social networks, focus is placed only on the community evolutionary outliers that show remarkable difference from the main bodies of their communities and sharp change of their membership role within the communities. A definition on such type of local and global outliers is given, and an algorithm on detection such kind of outliers is proposed in this paper. Experimental results on both synthetic and real datasets show that the proposed approach is highly effective in discovering interesting evolutionary community outliers.