Summary
Graphs can change a lot depending on time or place, and that makes it hard for computers to correctly understand new graphs that look different from those they learned on. The authors studied whether combining two ways of learning graph patterns—one that learns from labeled examples and another that learns from the graph itself without labels—helps computers do better on these new, different graphs. They created two methods: one that learns both kinds of information together, and another that combines the two separately at prediction time. Their tests show that using both kinds of learning leads to better recognition of nodes in new graph situations than just supervised learning alone.
What this means in practice
- •For social media engineers: Improve detection and labeling of users or content when social networks change over time or across regions by combining supervised and self-supervised graph learning.
- •For cybersecurity analysts: Enhance identification of malicious nodes in communication or transaction graphs that evolve or differ by network, using dual representation prediction methods.
Authors
Qingying Hao, Zikang Chen, Chuxuan Hu, Jinyuan Jia, Bo Li, Gang Wang, Carl Gunter
Abstract
Out-of-distribution (OOD) generalization remains challenging for graph neural networks (GNNs), as graph distributions can vary substantially across time and domains. Supervised and self-supervised graph representation learning are guided by distinct objectives and offer different perspectives on graph representations. In this work, we study whether self-supervised representations (SSL) can provide complementary signals to improve supervised OOD node classification. We develop two backbone-agnostic frameworks that exploit such information at different stages of learning and prediction. Co-Train jointly learns supervised and SSL representations and adaptively integrates them during training, while Dual-Space Retrieval performs non-parametric prediction in the two representation spaces and combines their predictions through confidence-aware fusion at inference time. The supervised and SSL encoders are separately parameterized and need not share the same GNN architecture. We evaluate multiple GNN backbones and two distinct SSL objectives, DGI and GRACE, on four graph benchmarks spanning temporal and cross-domain distribution shifts. Extensive experiments show that Co-Train consistently outperforms strong supervised OOD baselines, while Dual-Space Retrieval achieves competitive performance as a flexible non-parametric alternative. Results across different backbones and SSL objectives, together with representation analyses and ablations, demonstrate that SSL representations provide complementary information to supervised representations and can improve OOD node classification across diverse settings.