基于复发瞬间链接间隔的动态网络社区发现

Community Detection of Dynamic Networks Based on the Gaps of Recurrent Simultaneous Interactions

纵向网络数据是较为常见的复杂网络数据,也是目前网络数据分析的热点之一。随机块模型是网络社区发现的经典模型,但是该模型无法直接用于模拟纵向网络数据。基于随机块模型,引入半参数比例风险模型去分析纵向网络数据,并利用随机块模型来描述复发瞬间链接间隔。结合变分EM算法,采用两步估计来分别估计模型参数和非参数部分,通过不同场景下的模拟试验来验证所提议模型的优良性,最后利用法国小学生的社交网络数据进行了实证分析。模拟和实证结果表明,在统计计算的时效和参数或非参数估计的精度上,本文所提出的网络数据模型和统计分析方法比现存文献的模型和方法具有较好的优势。

Longitudinal network dataset is often from a complex network and is one of the hot topics in the statistical analysis of network data.The stochastic block model is a classic model of network community detection,but it cannot be directly used to fit longitudinal network data.Based on the stochastic block model,we propose a semi-parametric proportional risk model to analyze the longitudinal network data,in which the stochastic block model is applied to describe the recurrent instantaneous interactions.Combined with variational EM algorithm,two-step estimation was used to estimate model parameters and non-parametric parts respectively.Simulation studies in different scenarios were conducted to verify the merits of the proposed model.Finally,empirical analysis was carried out using the social network data of French primary school students.Simulation and demonstration show that the proposed network data model and statistical analysis method in this paper have some advantages than the existing models and methods in time consumpting of computation and the accuracy of parameter or non-parameter estimation.