AggreFact

该数据集整合了来自新闻源的微调模型生成的摘要的多个基准测试，它被用于分析传统摘要的事实性度量。其任务是对微调模型摘要的事实性进行评估。

This dataset integrates multiple benchmarks for summaries generated by fine-tuned models from news sources. It is employed to analyze factuality metrics for traditional summaries, with its primary objective being the evaluation of the factuality of summaries produced by fine-tuned models.

搜集汇总

背景与挑战

背景概述

AggreFact是一个专注于摘要生成中事实性错误的数据集，汇集了多个数据集的事实性错误标注，用于评估不同事实性度量标准的性能。该数据集特别关注了不同摘要模型（包括基于ChatGPT的模型）在事实性检测上的表现差异，并提供了详细的错误类型分析和性能比较。