DocRED (Document-Level Relation Extraction Dataset) is a relation extraction dataset constructed from Wikipedia and Wikidata. Each document in the dataset is human-annotated with named entity mentions, coreference information, intra- and inter-sentence relations, and supporting evidence.