Abstract:
|
Dynamic network modeling provides an emerging statistical technique to various real-world applications. It is a fundamental research question to detect the communities in large scale dynamic networks. However, due to significant computational challenges and difficulties in modeling communities, there is little progress in finding communities in dynamic networks. We present a novel model-based clustering framework for dynamic networks based on the exponential-family random graph models. We propose an effective model selection criterion to choose the number of communities. By using variational methods and MM techniques, we propose an efficient generalized variational expectation-maximization algorithm to solve approximate maximum likelihood estimates. Our method is demonstrated in an empirical application to the dynamic collaboration network data of a large northeastern research university. Our results provide insights about how different researchers work with their collaborators.
|