An End-to-end Framework for Graph Clustering: Semantic Fusion of Graph Convolutional Network and Variational Auto-encoder

Attribute graph clustering is a fundamental and challenging task in graph data mining, requiring the adequate utilization of both node attributes and graph structure. Recently, a series of graph clustering methods have been proposed, integrating both graph convolutional networks (GCNs) and autoencoders to capture structural information and node attributes, respectively. However, most existing methods either employ a complicated summation mechanism in the unaligned representation space from GCN and auto-encoder, or select a certain module (e.g. results from GCN) as a biased target selftraining signal. To address this, we propose a new end-to-end graph clustering framework that integrates a GCN and a variational autoencoder (VAE) with a more efficient and reasonable fusion mechanism in the semantic level. To better supervise the clustering, we select a high-confidence set of nodes based on the consensus of two encoders, further boosting the performances. Extensive experiments on multiple benchmark datasets demonstrate that our framework significantly outperforms state-of-the-art baselines, highlighting its efficacy in graph clustering.