Skip to Content

GraphWorld: Fake Graphs Bring Real Insights for GNNs

Posted on One min read
Accuracy curves obtained using GraphWorld.
GraphWorld allow us to explore the whole world of graphs for benchmarking. In comparisons classic datasets (dots) occupy only a small portion of the space.

paper · pdf · blogpost · code

TL;DR

Synthetic graph benchmarking for Graph Neural Networks.

We provide GraphWorld, a graph sampling and GNN training procedure which is capable of testing state-ofthe-art GNNs on task datasets beyond the scope of any existing benchmarks. We use GraphWorld to conduct large-scale experimental study on over 1 million graph datasets for each of three GNN tasks – node classification, link prediction, and graph property prediction. We provide a novel method to explore the GNN model performance across all locations in the graph worlds that we generate.