Unsupervised learning for community detection in attributed networks based on graph convolutional network

Abstract

Community detection has emerged during the last decade as one of the most challenging problems in network science, which has been revisited with network representation learning recently and has attracted considerable attention. Many approaches have been proposed in recent years, including the latest methods based on graph convolutional network (GCN). Here, we propose a new network representation learning method based on GCN for community detection in attributed networks without prior label information. Inspired by the message pass mechanism of GCN and the local self-organizing property of community structure, we integrate a label sampling model and GCN into an unsupervised learning framework to uncover underlying community structures by fusing topology and attribute information. The label sampling model constructs a balanced training set by structural center location and neighbor node expansion to train the GCN. The experiments on various real-world networks give a comparison view to evaluate the proposed method. The experimental results demonstrate the proposed method performs more efficiently with a comparative performance over current state-of-the-art community detection algorithms.