Datasets Knowledge Embedding: WN18RR, FB15K-237, and YAGO3-10 Benchmarks

This repository aggregates standard knowledge graph benchmarks including WN18RR, FB15K-237, and YAGO3-10 for knowledge embedding research. Curated by simonepri and last updated in March 2020, the collection provides a centralized source for relational triples used in link prediction. It specifically includes filtered versions of datasets designed to prevent test set leakage.

Use Cases

• Link prediction using head and tail entity triples to predict missing relations
• Knowledge graph completion benchmarking across FB15K and WN18 variants
• Entity representation learning for downstream natural language processing tasks

Strengths

• Includes 5+ standard benchmark datasets for knowledge embedding
• MIT licensed for open research and commercial use
• Provides filtered versions like FB15K-237 to prevent evaluation bias

Limitations

• Temporal staleness as the repository has not been updated since 2020
• Lack of metadata regarding the specific entity and relation ID mappings used

Provenance

Source

Aggregated from WordNet, Freebase, and YAGO research benchmarks

Collection Method

Aggregated from existing research publications and repositories

Freshness

Last updated in March 2020; reflects the standard benchmarks used in knowledge embedding research up to that period.

Geography

Global