Abstract:
We propose different extensions to Dynamic Stochastic Block models (DSBM). First, a flexible DSBM for multi-layer networks. We show that our model provides a unified framework including community detection and Gravity equation literature in the context of international trade. Second, in a one-layer framework, we present a DSBM with infinite communities to make inference on the number of communities using Bayesian non-parametric techniques. This second contribution is applied to real network data on international financial flows, which allows us to find evidence of a core-periphery structure and different effects of uncertainty depending on community membership. Third, given the high heterogeneity in real network data affecting the performance of community detection methods, we extend the infinite community DSBM by controlling for different sources of heterogeneity produced by observable and non-observable factors and then applied to the global international trade network.
