A navigation-ready, identifier-encoded packaging of the original MetaQA movie-domain knowledge graph question answering benchmark introduced by Zhang et al. The release contains 43,234 entities, 9 relations, 134,741 triples, and 407,513 questions across 1–3 hops. Explicit source and answer entities are materialized for graph navigation, while local opaque EID/RID identifiers provide a compact machine-readable graph representation.
The EID/RID encoding and file organization do not change the provenance or authorship of MetaQA; the underlying dataset remains attributed to its original creators.
| Name | Size | Link |
|---|---|---|
| README.md | — | |
| LICENSE | — | |
| croissant.json | — | |
| SHA256SUMS.txt | — | |
| kg/triplets.txt | 2.43 MB | |
| metadata/node_data.csv | 975 KB | |
| metadata/relation_data.csv | 160 B | |
| qa/metaqa_1hop.csv | 25.0 MB | |
| qa/metaqa_2hop.csv | 55.1 MB | |
| qa/metaqa_3hop.csv | 84.3 MB | |
| qa/metaqa_nhop.csv | 164 MB |