HIT-TMG/Hansel

Hansel is a high-quality human-annotated Chinese entity linking (EL) dataset, used for testing Chinese EL systems' generalization ability to tail entities and emerging entities. The test set contains Few-shot (FS) and zero-shot (ZS) slices, has 10K examples and uses Wikidata as the corresponding knowledge base. The training and validation sets are from Wikipedia hyperlinks, useful for large-scale pretraining of Chinese EL systems.