Dynamic networks are a general language for describing time-evolving complexsystems, and discrete time network models provide an emerging statisticaltechnique for various applications. It is a fundamental research question todetect the community structure in time-evolving networks. However, due tosignificant computational challenges and difficulties in modeling communitiesof time-evolving networks, there is little progress in the current literatureto effectively find communities in time-evolving networks. In this work, wepropose a novel model-based clustering framework for time-evolving networksbased on discrete time exponential-family random graph models. To choose thenumber of communities, we use conditional likelihood to construct an effectivemodel selection criterion. Furthermore, we propose an efficient variationalexpectation-maximization (EM) algorithm to find approximate maximum likelihoodestimates of network parameters and mixing proportions. By using variationalmethods and minorization-maximization (MM) techniques, our method has appealingscalability for large-scale time-evolving networks. The power of our method isdemonstrated in simulation studies and empirical applications to internationaltrade networks and the collaboration networks of a large American researchuniversity.
展开▼
