DocRED-FE
收藏arXiv2023-03-21 更新2024-06-21 收录
下载链接：
https://github.com/PKU-TANGENT/DOCRED-FE
下载链接
链接失效反馈官方服务：
资源简介：
DocRED-FE是由北京大学多媒体信息处理国家重点实验室构建的大型文档级细粒度实体和关系抽取数据集。该数据集基于DocRED，通过重新设计包含11个粗粒度和119个细粒度的实体类型架构，并手动重新标注DocRED来改进。数据集包含2596个文档，总计516k个tokens，旨在通过提供更精细的实体类型信息来促进关系分类，并解决现有模型在实际应用场景中的局限性。
DocRED-FE is a large-scale document-level fine-grained entity and relation extraction dataset constructed by the State Key Laboratory of Multimedia Information Processing, Peking University. It is improved based on DocRED by redesigning the entity type taxonomy consisting of 11 coarse-grained and 119 fine-grained entity types, and manually re-annotating the original DocRED dataset. The dataset contains 2596 documents with a total of 516k tokens, aiming to facilitate relation classification by providing more fine-grained entity type information and addressing the limitations of existing models in real-world application scenarios.
提供机构：
北京大学多媒体信息处理国家重点实验室
创建时间：
2023-03-20
