Revisit-DocRED

该数据集名为Revisit-DocRED，是对DocRED数据集的又一次改进，旨在解决不完全标注、共指错误和不一致逻辑等问题。与Re-DocRED相似，它的目标是为文档级关系提取任务提供一个更加可靠的基准，以便对DocRE模型进行训练和评估。The dataset, named Revisit-DocRED, is another improvement over the DocRED dataset, aiming to address issues such as incomplete annotation, coreference errors, and inconsistent logic. Similar to Re-DocRED, its objective is to provide a more reliable benchmark for the document-level relation extraction task, so as to facilitate the training and evaluation of DocRE models.