MyloBishop/reverse_summarization_dataset

该数据集是从两个不同的HuggingFace数据集转换而来，分别标记为data和wiki。wiki数据集进一步分为description和no description两部分，分别代表更干净、更受欢迎的文章和不太干净、不太受欢迎的文章。数据集由parquet文件组成，包含`instruction, response, source`三列。此外，还提供了用于生成摘要和从摘要恢复原文的提示模板。

原始信息汇总

数据集概述

数据来源

• data: 来自jordiclive/scored_summarization_datasets
• wiki: 来自jordiclive/wikipedia-summary-dataset，分为"description"和"no description"两部分。

数据结构

数据集包含parquet文件，文件中的列包括：instruction, response, source。

数据内容

• full_to_summary (fts): 包含以下提示语，用于引导指令：

  • "Summarize the following text: {}"
  • "Make a summary of the following text: {}"
  • "Provide a summary of the following text: {}"
  • "Change the following text into a summary: {}"
  • "Create a summary of the following text: {}"
  • "Give a brief overview of the following text: {}"
  • "Condense the following text into a summary: {}"
  • "Provide a condensed version of the following text: {}"
  • "Make a brief summary of the following text: {}"
  • "Create a condensed overview of the following text: {}"

• summary_to_full (stf): 包含以下提示语，用于引导指令：

  • "Write the original text for the following summary: {}"
  • "Write the full text for the following summary: {}"
  • "Provide the inputted source that provided the following summary: {}"
  • "Revert the following summary back into the original text: {}"
  • "Write a text that couldve provided the following summary: {}"
  • "Write the original text that generated the following summary: {}"
  • "Provide the full text for the following summary: {}"
  • "Create the inputted source that provided the following summary: {}"
  • "Write the original source that provided the following summary: {}"
  • "Convert the following summary back into the original text: {}"
  • "Provide a text that could have been the input for the following summary: {}"