Node4All: Learning Node Representation Beyond Datasets

Node representation learning on graphs has advanced rapidly, yet most methods still require per-dataset training and hyperparameter tuning, limiting their reusability across diverse graph datasets. We introduce Node4All, a node representation learner applicable to arbitrary graph datasets without dataset-specific optimization. Node4All is built on two ideas. First, we introduce the Channel Graph Transformer (CGT), which enables a single model to process arbitrary graph datasets. Second, we propose a self-supervised learning framework on synthetic graphs that enables generalization across arbitrary datasets. We evaluate Node4All on node classification across 25 benchmarks against 21 baselines. Despite using a single fixed model without per-dataset tuning, Node4All ranks 5th overall. With an appropriate predictor, it also supports one-shot and in-context learning, outperforming recent graph foundation models. These results show that Node4All is not only reusable across arbitrary graph datasets but also effective in practice. Code and model checkpoints are available at https://anonymous.4open.science/r/node4all-A3F1.