OGB: Standardized Benchmarks for Node, Link, and Graph Machine Learning

Open Graph Benchmark (OGB) is a collection of diverse benchmark datasets for graph machine learning developed by the SNAP group at Stanford University. Updated through May 2025, it provides standardized data loaders and evaluators for node, link, and graph-level prediction tasks across multiple domains.

Use Cases

• Node property prediction using node features and adjacency matrices
• Link prediction to forecast missing connections between graph entities
• Graph-level classification for molecular property prediction

Strengths

• Standardized evaluation protocols for reproducible research
• Covers three distinct scales: node-level, link-level, and graph-level tasks
• MIT licensed for open research and commercial use

Limitations

• Specific column schemas vary significantly between individual benchmarks in the suite
• High computational and memory requirements for the 'large-scale' subset of graphs

Provenance

Source

Stanford Network Analysis Project (SNAP)

Collection Method

Aggregated from various primary sources and curated into a standardized format

Freshness

Maintained and updated as of May 2025.