biu-nlp/Controlled-Text-Reduction-dataset

该数据集包含受控文本缩减的三元组——文档-摘要对，以及文档中覆盖摘要的文本片段。任务输入是带有预选片段（“高亮”）的文档，输出是覆盖所有且仅覆盖高亮内容的文本。数据集格式包括`doc_text`（输入文本）、`summary_text`（输出文本）和`highlight_spans`（输入文本中导致输出文本的片段）。数据集分为两个子集：`DUC-2001-2002`（进一步分为训练、验证和测试集）和`CNN-DM`（只有一个分割）。This dataset contains triples for controlled text reduction — document-summary pairs, along with text spans in the documents that cover the summaries. The task input is a document with pre-selected spans (“highlights”), and the output is text that covers all and only the highlighted content. The dataset format includes `doc_text` (input text), `summary_text` (output text), and `highlight_spans` (the spans in the input text that give rise to the output text). The dataset is divided into two subsets: `DUC-2001-2002` (further split into training, validation, and test sets) and `CNN-DM` (with only one split).

原始信息汇总

数据集概述

数据集名称

Controlled Text Reduction

数据集内容

包含文档-摘要对及其对应的文档中涵盖摘要内容的文本范围（highlight_spans）。

数据集结构

• doc_text：输入文本。
• summary_text：输出文本。
• highlight_spans：输入文本中的范围，这些范围导致输出文本。

数据集示例

json {doc_text: The motion picture industrys most coveted award...with 32., summary_text: The Oscar, created 60 years ago by MGM...awarded person (32)., highlight_spans:[[0, 48], [50, 55], [57, 81], [184, 247], ..., [953, 975], [1033, 1081]]}

数据集子集

1. DUC-2001-2002：分为训练集、验证集和测试集。
2. CNN-DM：单一分割。

引用信息

若使用此数据集，请引用以下论文：

@misc{https://doi.org/10.48550/arxiv.2210.13449, doi = {10.48550/ARXIV.2210.13449}, url = {https://arxiv.org/abs/2210.13449}, author = {Slobodkin, Aviv and Roit, Paul and Hirsch, Eran and Ernst, Ori and Dagan, Ido}, keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {Controlled Text Reduction}, publisher = {arXiv}, year = {2022}, copyright = {Creative Commons Zero v1.0 Universal} }