English

HCL: Improving Graph Representation with Hierarchical Contrastive Learning

Machine Learning 2022-10-24 v1 Artificial Intelligence

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.

Keywords

Cite

@article{arxiv.2210.12020,
  title  = {HCL: Improving Graph Representation with Hierarchical Contrastive Learning},
  author = {Jun Wang and Weixun Li and Changyu Hou and Xin Tang and Yixuan Qiao and Rui Fang and Pengyong Li and Peng Gao and Guotong Xie},
  journal= {arXiv preprint arXiv:2210.12020},
  year   = {2022}
}

Comments

published at The 21st International Semantic Web Conference ( ISWC 2022 )

R2 v1 2026-06-28T04:11:19.670Z