Title: A dynamic degree and strength corrected stochastic block model with infinite communities
Authors: Ovielt Antonio Baltodano Lopez - Ca' Foscari University (Italy) [presenting]
Roberto Casarin - University Ca' Foscari of Venice (Italy)
Mauro Costantini - University of L'aquila (Italy)
Abstract: The high heterogeneity in real network data affects the performance of community detection methods from a modeling and computational perspective. We propose a dynamic stochastic block model with infinite communities that allows making inferences on the number of communities using Bayesian nonparametric techniques after controlling for degree and strength heterogeneity produced by observable and unobservable factors. We cope with the poor mixing of the number of communities by using an MCMC that combines the forward filtering backward sampling and the merge-split approaches within an adaptive framework. The application of this model to the effect of COVID on the global international trade network shows the complexity of trends experienced during the pandemic, and it identifies cases that resulted in negative consequences and others in an increase of trade opportunities.