A Generative Node-attribute Network Model for Detecting Generalized Structure

TL;DR

This paper introduces GNAN, a generative model that integrates node topology and attributes to detect various network structures beyond communities, such as bipartite and core-periphery, with improved accuracy and interpretability.

Contribution

The paper presents a novel generative model (GNAN) that detects multiple types of network structures using combined topological and attribute information, surpassing existing methods.

Findings

GNAN effectively detects diverse network structures.

The model improves community detection accuracy.

It demonstrates semantic interpretability of detected structures.

Abstract

Exploring meaningful structural regularities embedded in networks is a key to understanding and analyzing the structure and function of a network. The node-attribute information can help improve such understanding and analysis. However, most of the existing methods focus on detecting traditional communities, i.e., groupings of nodes with dense internal connections and sparse external ones. In this paper, based on the connectivity behavior of nodes and homogeneity of attributes, we propose a principle model (named GNAN), which can generate both topology information and attribute information. The new model can detect not only community structure, but also a range of other types of structure in networks, such as bipartite structure, core-periphery structure, and their mixture structure, which are collectively referred to as generalized structure. The proposed model that combines topological information and node-attribute information can detect communities more accurately than the model that only uses topology information. The dependency between attributes and communities can be automatically learned by our model and thus we can ignore the attributes that do not contain useful information. The model parameters are inferred by using the expectation-maximization algorithm. And a case study is provided to show the ability of our model in the semantic interpretability of communities. Experiments on both synthetic and real-world networks show that the new model is competitive with other state-of-the-art models.