Robust Causal Graph Representation Learning against Confounding Effects

Authors

  • Hang Gao Science and Technology on Integrated Information System Laboratory, Institute of Software Chinese Academy of Sciences University of Chinese Academy of Sciences
  • Jiangmeng Li Science and Technology on Integrated Information System Laboratory, Institute of Software Chinese Academy of Sciences University of Chinese Academy of Sciences
  • Wenwen Qiang Science and Technology on Integrated Information System Laboratory, Institute of Software Chinese Academy of Sciences University of Chinese Academy of Sciences
  • Lingyu Si Science and Technology on Integrated Information System Laboratory, Institute of Software Chinese Academy of Sciences University of Chinese Academy of Sciences
  • Bing Xu China Communications Technology Information Group Co., Ltd.
  • Changwen Zheng Science and Technology on Integrated Information System Laboratory, Institute of Software Chinese Academy of Sciences
  • Fuchun Sun Tsinghua University

DOI:

https://doi.org/10.1609/aaai.v37i6.25925

Keywords:

ML: Representation Learning, DMKM: Graph Mining, Social Network Analysis & Community Mining, ML: Causal Learning, ML: Deep Learning Theory, ML: Graph-based Machine Learning, ML: Learning Theory

Abstract

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. Experimental results demonstrate the effectiveness and generalization ability of RCGRL. Our codes are available at https://github.com/hang53/RCGRL.

Downloads

Published

2023-06-26

How to Cite

Gao, H., Li, J., Qiang, W., Si, L., Xu, B., Zheng, C., & Sun, F. (2023). Robust Causal Graph Representation Learning against Confounding Effects. Proceedings of the AAAI Conference on Artificial Intelligence, 37(6), 7624-7632. https://doi.org/10.1609/aaai.v37i6.25925

Issue

Section

AAAI Technical Track on Machine Learning I