新华深度报道长文档数据集

数据集为纯文本，计划汇聚50万篇深度报道，平均4000字，合计20亿字。每篇文章标注篇章结构（导语、正文、结语）、观点句、情感极性、关键实体、事件时间线等6类标签；并生成100字AI摘要。整体压缩后数据规模约0.6TB，支持按主题、人物、地域、时间等12个维度聚类。

This is a pure-text dataset intended to aggregate 500,000 in-depth reports, with an average length of 4,000 words per article and a total of 2 billion words. Each article is annotated with six types of labels including textual structure (lead paragraph, main body, concluding paragraph), opinion sentences, sentiment polarity, key entities, and event timelines, and is paired with a 100-word AI-generated summary. After overall compression, the total data size is approximately 0.6 TB, and it supports clustering across 12 dimensions such as topic, person, region, time and others.

搜集汇总

背景与挑战

背景概述

该数据集汇集了50万篇深度报道文章，每篇平均4000字，总计20亿字，每篇文章均标注了篇章结构、观点句、情感极性等6类标签，并自动生成100字AI摘要。数据集支持按主题、人物、地域和时间等12个维度进行聚类分析，压缩后数据规模约为0.6TB，适用于大规模文本分析和自然语言处理任务。