Graph autoencoder (GAE) for community detection in social networks

Community detection in networks is essential for understanding the underlying structure and relationships among nodes. Existing methods for community detection often face challenges in capturing the intricate structure and spatial proximity of nodes in real-world networks. This paper proposes a graph autoencoder (GAE)-based method for community detection in social networks. It consists of three steps for extracting communities in the real-world network. The first one is a matrix reconstruction process that refines the original network structure. In this step, the computation of the most influential nodes and the spatial proximity of nodes are integrated to enhance the representation of spatial proximity. In the second step, the spatial features are extracted that leverages the reconstructed matrix to generate a low-dimensional graph subspace. In the third step, K-means clustering is applied on extracted feature matrix. That leads more coherent and meaningful communities. The performance of the proposed method is evaluated on eight real-world network datasets. The results demonstrate that the proposed method outperforms existing approaches in terms of modularity and NMI, consistently identifying more distinct and meaningful communities across diverse network sizes.