Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HyperGCL: Multi-Modal Graph Contrastive Learning via Learnable Hypergraph Views

Khaled Mohammed Saifuddin, Shihao Ji, Esra Akbas

TL;DR

HyperGCL tackles the limitations of conventional Graph Contrastive Learning by introducing three multimodal hypergraph views—attribute-driven, local-structure, and global-structure—learned through adaptive Gumbel-Softmax augmentation and encoded with view-specific HyGAN/SHyGAN networks. It advances a topology-guided NetCL loss that defines positives and negatives using network structure and offers selective negative sampling (distance-based or similarity-based) to reduce computation. The combined framework achieves state-of-the-art node classification performance across five benchmark datasets, with ablations confirming the critical roles of each view, adaptive augmentation, NetCL, and SHyGAN-derived structural encodings. Overall, HyperGCL demonstrates the value of integrating higher-order multimodal graph representations and topology-aware objectives for robust, scalable graph representation learning.

Abstract

Recent advancements in Graph Contrastive Learning (GCL) have demonstrated remarkable effectiveness in improving graph representations. However, relying on predefined augmentations (e.g., node dropping, edge perturbation, attribute masking) may result in the loss of task-relevant information and a lack of adaptability to diverse input data. Furthermore, the selection of negative samples remains rarely explored. In this paper, we introduce HyperGCL, a novel multimodal GCL framework from a hypergraph perspective. HyperGCL constructs three distinct hypergraph views by jointly utilizing the input graph's structure and attributes, enabling a comprehensive integration of multiple modalities in contrastive learning. A learnable adaptive topology augmentation technique enhances these views by preserving important relations and filtering out noise. View-specific encoders capture essential characteristics from each view, while a network-aware contrastive loss leverages the underlying topology to define positive and negative samples effectively. Extensive experiments on benchmark datasets demonstrate that HyperGCL achieves state-of-the-art node classification performance.

HyperGCL: Multi-Modal Graph Contrastive Learning via Learnable Hypergraph Views

TL;DR

HyperGCL tackles the limitations of conventional Graph Contrastive Learning by introducing three multimodal hypergraph views—attribute-driven, local-structure, and global-structure—learned through adaptive Gumbel-Softmax augmentation and encoded with view-specific HyGAN/SHyGAN networks. It advances a topology-guided NetCL loss that defines positives and negatives using network structure and offers selective negative sampling (distance-based or similarity-based) to reduce computation. The combined framework achieves state-of-the-art node classification performance across five benchmark datasets, with ablations confirming the critical roles of each view, adaptive augmentation, NetCL, and SHyGAN-derived structural encodings. Overall, HyperGCL demonstrates the value of integrating higher-order multimodal graph representations and topology-aware objectives for robust, scalable graph representation learning.

Abstract

Recent advancements in Graph Contrastive Learning (GCL) have demonstrated remarkable effectiveness in improving graph representations. However, relying on predefined augmentations (e.g., node dropping, edge perturbation, attribute masking) may result in the loss of task-relevant information and a lack of adaptability to diverse input data. Furthermore, the selection of negative samples remains rarely explored. In this paper, we introduce HyperGCL, a novel multimodal GCL framework from a hypergraph perspective. HyperGCL constructs three distinct hypergraph views by jointly utilizing the input graph's structure and attributes, enabling a comprehensive integration of multiple modalities in contrastive learning. A learnable adaptive topology augmentation technique enhances these views by preserving important relations and filtering out noise. View-specific encoders capture essential characteristics from each view, while a network-aware contrastive loss leverages the underlying topology to define positive and negative samples effectively. Extensive experiments on benchmark datasets demonstrate that HyperGCL achieves state-of-the-art node classification performance.

Paper Structure

This paper contains 16 sections, 17 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Multimodal Hypergraph View Generation and Adaptive Augmentation
  5. Attribute-driven Hypergraph View ($\mathcal{H}^{a}$)
  6. Structure-driven Hypergraph Views
  7. Adaptive View Augmentation
  8. View-Specific Hypergraph Encoder
  9. Node's Structural Feature Encoding
  10. Topology-Guided Attention
  11. Network-Aware Contrastive Loss (NetCL)
  12. Experiment
  13. Experimental Setup
  14. Performance Comparison
  15. Ablation Study
  16. ...and 1 more sections

Figures (2)

  • Figure 1: System architecture of HyperGCL. After constructing three different hypergraph views from the input graph and node attributes, we exploit a learnable view augmentation technique to generate adaptive views. View-specific encoders are used to learn each view and finally, a network-aware contrastive loss is used with a supervised loss to train the model.
  • Figure 2: The performance (accuracy %) of HyperGCL with different numbers of global nodes ($n_g$) in $\mathcal{H}^g$.