Dynamic clustering of network data: a hybrid maximum likelihood approach

Dynamic Stochastic Block Models (SBMs) represent an attractive field of research as they provide a flexible modeling tool for the dynamic clustering of network data. However, full maximum likelihood (ML) for this class of models is not a viable estimating strategy due to the intractability of the likelihood function: variational inference represents quite a standard alternative in the frequentist framework. However, despite its simplicity, it may lead to non-optimal estimators and may suffer from local maxima solutions. We extend the hybrid ML approach developed in the context of static SBMs to deal with dynamic networks also considering both weighted and unweighted relations as well as nodal attributes which may potentially affect the block structure.