community-datasets/qa_zre
收藏资源简介：
QaZre数据集是一个将关系抽取任务简化为阅读理解问题的数据集。它包含训练、验证和测试三个部分，分别有8400000、6000和120000个样本。数据集的字段包括关系、问题、主题、上下文和答案。该数据集主要用于问答任务，特别是零样本关系抽取。数据集的创建者未提及，且许可证信息未知。
The QaZre dataset is a benchmark that simplifies the relation extraction task into reading comprehension-style questions. It consists of three splits: training, validation, and test, with 8,400,000, 6,000, and 120,000 samples respectively. The fields of this dataset include relation, question, subject, context, and answer. This dataset is primarily intended for question answering tasks, especially zero-shot relation extraction. The creator of this dataset is not specified, and the license information is unknown.
数据集概述
数据集基本信息
- 数据集名称: QaZre
- 语言: 英语
- 许可证: 未知
- 多语言性: 单语种
- 数据集大小: 1M<n<10M
- 源数据: 原始数据
- 任务类别: 问答
- 标签: zero-shot-relation-extraction
数据集结构
特征
- relation: 字符串类型
- question: 字符串类型
- subject: 字符串类型
- context: 字符串类型
- answers: 字符串序列
数据分割
- 训练集: 8400000 条数据
- 验证集: 6000 条数据
- 测试集: 120000 条数据
数据实例
json { "answers": [], "context": "answer", "question": "What is XXX in this question?", "relation": "relation_name", "subject": "Some entity Here is a bit of context which will explain the question in some way" }
引用信息
@inproceedings{levy-etal-2017-zero, title = "Zero-Shot Relation Extraction via Reading Comprehension", author = "Levy, Omer and Seo, Minjoon and Choi, Eunsol and Zettlemoyer, Luke", booktitle = "Proceedings of the 21st Conference on Computational Natural Language Learning ({C}o{NLL} 2017)", month = aug, year = "2017", address = "Vancouver, Canada", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/K17-1034", doi = "10.18653/v1/K17-1034", pages = "333--342", }
