Profile-pseudo likelihood methods for community detection of multilayer stochastic block models

The multilayer stochastic block model is one of the fundamental models in multilayer networks and is often used to represent multiple types of relations between different individuals. In this paper, we extend the profile-pseudo likelihood method for the single-layer stochastic block model to the case of the multilayer stochastic block model. Specifically, by assuming all network layers have identical community membership labels, we investigate the multilayer stochastic block model with a common community structure. In this paper, we develop a profile-pseudo likelihood algorithm to fit a multilayer stochastic block model and estimate the community label. Meantime, we prove that the algorithm has convergence guarantee and that the estimated community label is strongly consistent. Further, for estimating the number of communities K$$ K $$, we extend the corrected Bayesian information criterion to multilayer stochastic block models. We also extend this algorithm to fit the multilayer degree-corrected stochastic block model. Both simulation studies and real-world data examples indicate that the proposed method works well.

Automated summary

In this paper, the profile-pseudo likelihood method is extended from the single-layer stochastic block model to the multilayer stochastic block model, with the assumption of identical community membership labels across network layers. The proposed algorithm is proven to have convergence guarantee and produce strongly consistent estimated community labels. The method is further applied to the multilayer degree-corrected stochastic block model, and both simulation studies and real-world data examples show its effectiveness.