GraphBench: Next-generation graph learning benchmarking

Researchers have introduced GraphBench, a comprehensive benchmark suite designed to address the fragmentation and inconsistency in current machine learning on graphs practices. Despite significant progress in areas like molecular property prediction and chip design, existing benchmarks often rely on narrow datasets and inconsistent evaluation protocols, which hinder reproducibility and broader scientific advancement. The rise of graph foundation models has further exposed these limitations. GraphBench spans diverse real-world domains and task settings, including node-level, edge-level, graph-level, and generative tasks. It provides standardized evaluation protocols with consistent dataset splits and metrics for assessing out-of-distribution generalization, alongside a unified hyperparameter-tuning framework. The study evaluates GraphBench using recent message-passing neural networks and graph transformer models to establish principled baselines for future research. This initiative aims to facilitate more rigorous and comparable evaluations in the field of graph learning, supporting the development of more robust and generalizable models. Further details are available at www.graphbench.io.