A large-scale benchmark for evaluating the expressive power of Graph Neural Networks (GNNs) across 16 fundamental graph properties. This dataset accompanies the paper "Systematic Property-Driven Evaluation of
GNN Expressiveness".
The benchmark contains 352 graph-classification datasets organized into two families:
GraphRandom (176 datasets): graphs that either satisfy or randomly violate a given property.
GraphPerturb (176 datasets):… See the full description on the dataset page:
https://huggingface.co/datasets/anonymousPaper5674/Property-Driven-GNN.
