ChemDisGene

ChemDisGene is a new dataset for training and evaluating multi-type multi-label biomedical relation extraction models. It contains a large automatically derived corpus with noisy relationship labels and a cleaner manually curated dataset of 523 abstracts.