TGB: Temporal Graph Benchmark Datasets in RelBench Format

RelBench hosts the Temporal Graph Benchmark datasets converted to its manifest format. The repository contains multiple sub-datasets, each with a self-describing manifest, parquet tables, and task-specific labels. The data was last updated on 2026-06-12.

Use Cases

• Benchmarking temporal graph neural networks based on the structured relational tasks.
• Evaluating link prediction models on dynamic graphs using the provided external labels and splits.
• Testing relational learning algorithms on standardized temporal graph datasets with official negative samples.

Strengths

• Datasets are formatted according to the standardized RelBench manifest specification.
• Each task includes official labels, splits, and negative samples produced by the Temporal Graph Benchmark.
• Data was last updated on 2026-06-12, indicating recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The specific scale (rows, size) of individual datasets is unknown from the provided metadata.

Provenance

Source

Temporal Graph Benchmark (TGB), exported by RelBench.

Collection Method

Converted and exported to the RelBench manifest format.

Freshness

Last updated 2026-06-12 07:58:42.