PGC-CSS: A parallel graph clustering framework with collaborative self-supervision

doi:10.1016/j.knosys.2025.114010

PGC-CSS: A parallel graph clustering framework with collaborative self-supervision

2025-07-03

0

PRE

AI

DOI：10.1016/j.knosys.2025.114010

Original

Original request for help

Save

Abstract

En 中文

Graph clustering is a fundamental yet challenging task in data analysis, aiming to partition nodes into coherent groups based on graph topology and node features. Recently, deep clustering methods based on graph convolutional networks (GCN) have achieved remarkable success. However, most existing approaches rely on a single learning pipeline built upon the original graph structure, which is often corrupted by noisy or unreliable edges. As a result, the clustering performance becomes unstable. To address the above issue, we propose a novel parallel graph clustering framework, termed PGC-CSS. Specifically, we introduce a graph refinement method that alleviates noise in the original graph by uncovering latent structural patterns from node embeddings. The refined graph is then integrated with the original graph in a parallel architecture, enabling robust and discriminative representation learning through the complementary modeling of multiple graph structures. Furthermore, we design a collaborative self-supervision module that combines distribution-based and pseudo-label-based self-supervision strategies to guide the model toward learning optimal clustering representations. Extensive experiments on six benchmark datasets demonstrate the effectiveness and superiority of PGC-CSS for deep attributed graph clustering.

PGC-CSS: A parallel graph clustering framework with collaborative self-supervision

Abstract

Journal

Organization