Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Catch Causal Signals from Edges for Label Imbalance in Graph Classification

Fengrui Zhang, Yujia Yin, Hongzong Li, Yifan Chen, Tianyi Qu

TL;DR

This work tackles label imbalance in graph classification by integrating edge features into a causal attention framework (ECAL). By distinguishing causal from noncausal components through edge-enhanced attention and a causal subgraph splitting mechanism, the method improves out-of-distribution generalization on molecular and bioinformatics datasets. The approach combines two EGAT variants, a three-term loss including KL and backdoor-like terms, and explicit edge-aware representations to yield more robust graph-level predictions. The results demonstrate significant gains over baselines, highlighting the importance of edge information for causal discovery in graphs and offering a practical path for robust graph classification in imbalanced settings.

Abstract

Despite significant advancements in causal research on graphs and its application to cracking label imbalance, the role of edge features in detecting the causal effects within graphs has been largely overlooked, leaving existing methods with untapped potential for further performance gains. In this paper, we enhance the causal attention mechanism through effectively leveraging edge information to disentangle the causal subgraph from the original graph, as well as further utilizing edge features to reshape graph representations. Capturing more comprehensive causal signals, our design leads to improved performance on graph classification tasks with label imbalance issues. We evaluate our approach on real-word datasets PTC, Tox21, and ogbg-molhiv, observing improvements over baselines. Overall, we highlight the importance of edge features in graph causal detection and provide a promising direction for addressing label imbalance challenges in graph-level tasks. The model implementation details and the codes are available on https://github.com/fengrui-z/ECAL

Catch Causal Signals from Edges for Label Imbalance in Graph Classification

TL;DR

This work tackles label imbalance in graph classification by integrating edge features into a causal attention framework (ECAL). By distinguishing causal from noncausal components through edge-enhanced attention and a causal subgraph splitting mechanism, the method improves out-of-distribution generalization on molecular and bioinformatics datasets. The approach combines two EGAT variants, a three-term loss including KL and backdoor-like terms, and explicit edge-aware representations to yield more robust graph-level predictions. The results demonstrate significant gains over baselines, highlighting the importance of edge information for causal discovery in graphs and offering a practical path for robust graph classification in imbalanced settings.

Abstract

Despite significant advancements in causal research on graphs and its application to cracking label imbalance, the role of edge features in detecting the causal effects within graphs has been largely overlooked, leaving existing methods with untapped potential for further performance gains. In this paper, we enhance the causal attention mechanism through effectively leveraging edge information to disentangle the causal subgraph from the original graph, as well as further utilizing edge features to reshape graph representations. Capturing more comprehensive causal signals, our design leads to improved performance on graph classification tasks with label imbalance issues. We evaluate our approach on real-word datasets PTC, Tox21, and ogbg-molhiv, observing improvements over baselines. Overall, we highlight the importance of edge features in graph causal detection and provide a promising direction for addressing label imbalance challenges in graph-level tasks. The model implementation details and the codes are available on https://github.com/fengrui-z/ECAL
Paper Structure (16 sections, 15 equations, 3 figures, 2 tables)

This paper contains 16 sections, 15 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Task formulation
  4. Causal discovery and causal attention mechanisms
  5. Methods
  6. Incorporating edge features into attention
  7. Graph causal representation with edge information
  8. Causal subgraph splitting
  9. Edge-featured causal and trivial representations
  10. Loss computation
  11. Experiment Results
  12. Baselines and main comparison
  13. Ablation studies
  14. Ablating causality detection
  15. Adding noise to ablate edge features
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Framework overview. The proposed model integrates causal attention mechanisms with edge feature incorporation, enhancing the generalization capabilities of GNNs in OOD scenarios. Best viewed zoom in.
  • Figure 2: The effect on OOD accuracy of removing causality detection loss terms, for both CAL-based and ECAL-based models.
  • Figure 3: OOD accuracy under different noise levels (proportion of permutation) and label imbalance levels on ogbg-molhiv.