Node Classification

Node classification predicts categories for entities such as users, papers, accounts, or neurons, using their attributes and relationships. Common approaches include graph neural networks with message passing, label propagation, graph transformers, and methods that separate homophilic from heterophilic connections; text-attributed graphs also use language models and graph foundation models. Work covers transductive and inductive prediction, as well as few-shot learning, class imbalance, noisy labels, distribution shift, and calibrated uncertainty. A central challenge is determining when added structure or pretrained models truly help: results can depend on graph type, data splits, and careful tuning, and strong conventional GNN baselines remain difficult to beat consistently.