WikiSum

该数据集包含了维基百科文章及其来源参考文献，主要用于多文档摘要任务。平均而言，每个输入包含525个段落，每个段落有70.1个词汇，目标摘要的平均长度为139.4个词汇。该数据集的规模为：训练实例1579360个，验证实例38144个，测试实例38205个，所涉及的任务是多文档摘要。

This dataset comprises Wikipedia articles and their corresponding source references, primarily intended for multi-document summarization tasks. On average, each input consists of 525 paragraphs, with each paragraph containing 70.1 words, and the average length of the target summaries is 139.4 words. The dataset includes 1,579,360 training instances, 38,144 validation instances, and 38,205 test instances, with the targeted task being multi-document summarization.

搜集汇总

背景与挑战

背景概述

WikiSum是一个用于生成维基百科文章的摘要数据集，基于论文《Generating Wikipedia by Summarizing Long Sequences》。它包含约230万篇维基百科文章及其引用内容，来源包括CommonCrawl（可重复但覆盖有限）和实时网络抓取（覆盖更广但内容可变），总URL数约9000万。数据集的输入是引用内容，输出是维基百科文章，用于训练摘要生成模型。