Inferring community structure in attributed hypergraphs using stochastic block models

Kazuki Nakajima,Takeaki Uno

2024-01-01

Abstract:Hypergraphs are a representation of complex systems involving interactions among more than two entities and allow to investigation of higher-order structure and dynamics in real-world complex systems. Community structure is a common property observed in empirical networks in various domains. Stochastic block models have been employed to investigate community structure in networks. Node attribute data, often accompanying network data, has been found to potentially enhance the learning of community structure in dyadic networks. In this study, we develop a statistical framework that incorporates node attribute data into the learning of community structure in a hypergraph, employing a stochastic block model. We demonstrate that our model, which we refer to as HyperNEO, enhances the learning of community structure in synthetic and empirical hypergraphs when node attributes are sufficiently associated with the communities. Furthermore, we found that applying a dimensionality reduction method, UMAP, to the learned representations obtained using stochastic block models, including our model, maps nodes into a two-dimensional vector space while largely preserving community structure in empirical hypergraphs. We expect that our framework will broaden the investigation and understanding of higher-order community structure in real-world complex systems.

Social and Information Networks,Machine Learning