Robust Causal Graph Representation Learning against Confounding Effects
This addresses a specific issue in graph representation learning for researchers and practitioners, offering a novel method to improve robustness against confounding effects, though it appears incremental as it builds on existing graph neural network frameworks.
The paper tackles the problem of confounding effects in graph neural networks, which cause models to underperform when using full graphs compared to pruned ones, and proposes Robust Causal Graph Representation Learning (RCGRL) to eliminate these confounders, achieving better prediction performance and generalization ability than state-of-the-art methods in experiments.
The prevailing graph neural network models have achieved significant progress in graph representation learning. However, in this paper, we uncover an ever-overlooked phenomenon: the pre-trained graph representation learning model tested with full graphs underperforms the model tested with well-pruned graphs. This observation reveals that there exist confounders in graphs, which may interfere with the model learning semantic information, and current graph representation learning methods have not eliminated their influence. To tackle this issue, we propose Robust Causal Graph Representation Learning (RCGRL) to learn robust graph representations against confounding effects. RCGRL introduces an active approach to generate instrumental variables under unconditional moment restrictions, which empowers the graph representation learning model to eliminate confounders, thereby capturing discriminative information that is causally related to downstream predictions. We offer theorems and proofs to guarantee the theoretical effectiveness of the proposed approach. Empirically, we conduct extensive experiments on a synthetic dataset and multiple benchmark datasets. The results demonstrate that compared with state-of-the-art methods, RCGRL achieves better prediction performance and generalization ability.