HCL: Improving Graph Representation with Hierarchical Contrastive Learning

TL;DR

This paper introduces HCL, a hierarchical contrastive learning framework for graphs that captures multi-scale information and improves representation quality through adaptive pooling and a multi-channel network, showing strong results across various tasks.

Contribution

HCL is the first to explicitly learn hierarchical graph representations using adaptive pooling and a multi-channel network, enhancing the capture of local and global graph features.

Findings

HCL outperforms existing methods on 12 benchmark datasets.

HCL effectively captures meaningful hierarchical graph features.

Visualizations confirm richer and more comprehensive graph representations.

Abstract

Contrastive learning has emerged as a powerful tool for graph representation learning. However, most contrastive learning methods learn features of graphs with fixed coarse-grained scale, which might underestimate either local or global information. To capture more hierarchical and richer representation, we propose a novel Hierarchical Contrastive Learning (HCL) framework that explicitly learns graph representation in a hierarchical manner. Specifically, HCL includes two key components: a novel adaptive Learning to Pool (L2Pool) method to construct more reasonable multi-scale graph topology for more comprehensive contrastive objective, a novel multi-channel pseudo-siamese network to further enable more expressive learning of mutual information within each scale. Comprehensive experimental results show HCL achieves competitive performance on 12 datasets involving node classification, node clustering and graph classification. In addition, the visualization of learned representation reveals that HCL successfully captures meaningful characteristics of graphs.