Clustering Time-Evolving Networks Using Temporal Exponential-Family Random Graph Models with Conditional Dyadic Independence and Dynamic Latent Blocks
提出一种基于时间指数族随机图模型和隐马尔可夫结构的统计聚类框架，用于探索时变网络的动态潜在块结构，并开发了变分EM算法进行参数估计和模型选择。
Model-based clustering of dynamic networks has emerged as an increasingly important research topic in statistical network analysis. Effectively and efficiently exploring the dynamic latent block structure of time-evolving networks is critical. We propose a statistical clustering framework based on temporal exponential-family random graph models (ERGMs) with conditional dyadic independence and a hidden Markov structure. These conditional independent temporal ERGMs allow for the specification of meaningful network features, while the hidden Markov structure helps infer the dynamic latent block structure. Additionally, we develop a variational expectation-maximization algorithm to approximate maximum likelihood estimation and present an effective model selection criterion, based on the integrated classification likelihood, to determine the optimal number of clusters. Finally, we demonstrate the numerical performance of our proposed method through extensive simulation studies and real-world applications.