Graph Augmentation for Cross Graph Domain Generalization

TL;DR

This paper introduces a novel graph structure augmentation method combining edge-dropping and clustering-based edge-adding to improve cross-graph domain generalization in node classification tasks, demonstrating state-of-the-art results.

Contribution

It proposes a new graph augmentation technique specifically designed for cross-graph domain generalization, addressing the limitations of existing training-focused methods.

Findings

Achieves state-of-the-art performance on out-of-distribution citation datasets.

Effectively removes noise edges to enhance invariant structure learning.

Improves GNN generalization through structure-preserving augmentations.

Abstract

Cross-graph node classification, utilizing the abundant labeled nodes from one graph to help classify unlabeled nodes in another graph, can be viewed as a domain generalization problem of graph neural networks (GNNs) due to the structure shift commonly appearing among various graphs. Nevertheless, current endeavors for cross-graph node classification mainly focus on model training. Data augmentation approaches, a simple and easy-to-implement domain generalization technique, remain under-explored. In this paper, we develop a new graph structure augmentation for the crossgraph domain generalization problem. Specifically, low-weight edgedropping is applied to remove potential noise edges that may hinder the generalization ability of GNNs, stimulating the GNNs to capture the essential invariant information underlying different structures. Meanwhile, clustering-based edge-adding is proposed to generate invariant structures based on the node features from the same distribution. Consequently, with these augmentation techniques, the GNNs can maintain the domain invariant structure information that can improve the generalization ability. The experiments on out-ofdistribution citation network datasets verify our method achieves state-of-the-art performance among conventional augmentations.