The KGC benchmark evaluates a model's ability to complete missing links in a knowledge graph.
Format: (subject, relation, object) triples
Splits: Train/Dev/Test
Task: Predict missing entity (either subject or object)
Construction: Extracted from entity-valued synsets and projected to (S, P, O) triples with careful cross-split collision handling… See the full description on the dataset page:
https://huggingface.co/datasets/openbmb/factnet-bench.
