Community analysis of interfirm networks with graph convolutional networks

Abstract

Community detection is one of the important tasks in complex network analysis, and detected community itself should be also remarkable to study. In this study, we analyze the properties of community, detected by Infomap, with graph convolutional networks. We study the Japanese inter-firm network researched by TOKYO SHOKO RESEARCH, LTD as the example of community analysis. The embedding expressions of community obtained through the variational graph auto-encoder reveal the relation of communities and the characteristics of industry and region in communities.