Baike2018qa

百科问答 (baike2018qa)，包含150万个预先过滤的高质量问答，每个类别一个。共有492个类别，其中434个类别的频率为10个或更多。

数据集分区: 数据被重复数据删除并分为三个部分。训练集: 142.5万; 验证集: 45,000; 测试集，数万，不可下载。

The baike2018qa encyclopedic question answering dataset contains 1.5 million pre-filtered high-quality question-answer pairs, with one pair per category. In total, there are 492 categories, and 434 of these categories include 10 or more question-answer pairs.

Dataset Partitioning: The dataset has been deduplicated and divided into three subsets. The training set includes 1,425,000 pairs, the validation set contains 45,000 pairs, and the test set consists of tens of thousands of pairs, which is not available for download.

搜集汇总

背景与挑战

背景概述

Baike2018qa是一个中文百科问答数据集，包含150万个高质量问答，覆盖492个类别，其中434个类别的频率达到10次以上。数据集已进行去重处理，并划分为训练集（142.5万）、验证集（4.5万）和测试集（数万，不可下载），适用于自然语言处理任务如问答系统训练。