Learning from Counterfactual Links for Link Prediction

Tong Zhao, Gang Liu, Daheng Wang, Wenhao Yu, Meng Jiang
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:26911-26926, 2022.

Abstract

Learning to predict missing links is important for many graph-based applications. Existing methods were designed to learn the association between observed graph structure and existence of link between a pair of nodes. However, the causal relationship between the two variables was largely ignored for learning to predict links on a graph. In this work, we visit this factor by asking a counterfactual question: "would the link still exist if the graph structure became different from observation?" Its answer, counterfactual links, will be able to augment the graph data for representation learning. To create these links, we employ causal models that consider the information (i.e., learned representations) of node pairs as context, global graph structural properties as treatment, and link existence as outcome. We propose a novel data augmentation-based link prediction method that creates counterfactual links and learns representations from both the observed and counterfactual links. Experiments on benchmark data show that our graph learning method achieves state-of-the-art performance on the task of link prediction.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-zhao22e, title = {Learning from Counterfactual Links for Link Prediction}, author = {Zhao, Tong and Liu, Gang and Wang, Daheng and Yu, Wenhao and Jiang, Meng}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {26911--26926}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/zhao22e/zhao22e.pdf}, url = {https://proceedings.mlr.press/v162/zhao22e.html}, abstract = {Learning to predict missing links is important for many graph-based applications. Existing methods were designed to learn the association between observed graph structure and existence of link between a pair of nodes. However, the causal relationship between the two variables was largely ignored for learning to predict links on a graph. In this work, we visit this factor by asking a counterfactual question: "would the link still exist if the graph structure became different from observation?" Its answer, counterfactual links, will be able to augment the graph data for representation learning. To create these links, we employ causal models that consider the information (i.e., learned representations) of node pairs as context, global graph structural properties as treatment, and link existence as outcome. We propose a novel data augmentation-based link prediction method that creates counterfactual links and learns representations from both the observed and counterfactual links. Experiments on benchmark data show that our graph learning method achieves state-of-the-art performance on the task of link prediction.} }
Endnote
%0 Conference Paper %T Learning from Counterfactual Links for Link Prediction %A Tong Zhao %A Gang Liu %A Daheng Wang %A Wenhao Yu %A Meng Jiang %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-zhao22e %I PMLR %P 26911--26926 %U https://proceedings.mlr.press/v162/zhao22e.html %V 162 %X Learning to predict missing links is important for many graph-based applications. Existing methods were designed to learn the association between observed graph structure and existence of link between a pair of nodes. However, the causal relationship between the two variables was largely ignored for learning to predict links on a graph. In this work, we visit this factor by asking a counterfactual question: "would the link still exist if the graph structure became different from observation?" Its answer, counterfactual links, will be able to augment the graph data for representation learning. To create these links, we employ causal models that consider the information (i.e., learned representations) of node pairs as context, global graph structural properties as treatment, and link existence as outcome. We propose a novel data augmentation-based link prediction method that creates counterfactual links and learns representations from both the observed and counterfactual links. Experiments on benchmark data show that our graph learning method achieves state-of-the-art performance on the task of link prediction.
APA
Zhao, T., Liu, G., Wang, D., Yu, W. & Jiang, M.. (2022). Learning from Counterfactual Links for Link Prediction. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:26911-26926 Available from https://proceedings.mlr.press/v162/zhao22e.html.

Related Material