A0809
Title: Bayesian community detection in assortative stochastic block model with an unknown number of communities
Authors: Martina Amongero - University of Torino (Italy) [presenting]
Pierpaolo De Blasi - University of Torino and Collegio Carlo Alberto (Italy)
Abstract: Available data in the form of networks is gaining increasing attention in modern research; examples include social networks, biological networks, and many others. A statistical problem of key interest is community detection, that is, to divide the nodes into strongly connected clusters with relatively weak cross-cluster connections. The stochastic block model (SBM) provides a well-suited generative process for explaining the formation of communities. By leveraging the SBM, researchers gain insights into the underlying structure of a network, uncovering interaction patterns that may not be apparent from raw data, and exploring network properties such as assortativity. In particular, an SBM is assortative when the probability of a connection is higher when nodes belong to the same rather than to different clusters. A recent line of work uses Bayesian non-parametric methods for the recovery of communities in classical SBM by placing a prior distribution on the number of clusters and estimating cluster assignments with collapsed Gibbs samplers. However, the development of an efficient Gibbs sampler for assortative SBM is still an open problem. The aim is to enforce the assortative property through the prior and to study its effect on community detection by implementing a Gibbs sampler for posterior inference. A simulation study based on a class of benchmark datasets is conducted to demonstrate the benefits of the assortative case over the standard one.
