Graph Fusion Based Autoencoder for Node Clustering

Ci Nie, Yujing Liu, Jian Wei, Guoqiu Wen

Abstract: Node clustering is widely applied on graph-structured data, as it groups similar nodes into the same subspace and disperses dissimilar nodes into distinct subspaces. However, existing methods still suffer from the following issues: 1) the homogeneous graph assumption often leads to negative effect; 2) previous methods ignore the inter-cluster relationship. To address these issues, this paper proposes Graph Fusion based Autoencoder for Node Clustering (GFANC). Specifically, an attribute graph is first constructed and then integrated with the topological graph to diminish the weight of heterogeneous edges, thereby alleviating their negative effects. Furthermore, both intra-cluster and inter-cluster relationships are taken into consideration to achieve well-defined clustering boundaries. Experimental results on three real datasets show that our method achieves the best results compared to SOTA methods.

External IDs:dblp:conf/adma/NieLWW24