310,079 triples across 237 relation types and 14,541 entities derived from the Freebase knowledge base. The dataset is structured for link prediction tasks, providing head entities, relations, and tail entities while specifically excluding redundant inverse relations found in its predecessor.
Use Cases
- Train knowledge graph embedding models like TransE or RotatE using the head, relation, and tail triples
- Evaluate link prediction performance by predicting the missing tail entity given a head entity and relation
- Benchmark relation extraction algorithms on a dataset specifically designed to prevent inverse relation leakage
Strengths
- Contains 310,079 triples representing knowledge graph facts
- Features 237 distinct relation types reduced from the original 1,345 in FB15k
- Includes 14,541 unique entities sourced from Freebase
- Structured as triples consisting of head entity, relation, and tail entity