DISLab/FeedSum
收藏Hugging Face2025-01-25 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/DISLab/FeedSum
下载链接
链接失效反馈官方服务：
资源简介：
FeedSum是一个大规模的用于文本摘要的LLM生成反馈数据集，包含多样化的文档和摘要对，以及针对摘要质量的四种不同配置的LLM生成反馈。这些反馈配置包括低质量单一维度粗糙粒度评估、高质量单一维度粗糙粒度评估、高质量多维度粗糙粒度评估和高质量多维度细致粒度评估。数据集还包括人工参考摘要和利用GPT4-turbo提取的文档关键事实。
FeedSum is a large-scale LLM-generated feedback dataset for text summarization, containing diverse document and summary pairs along with LLM-generated feedback on summary quality in four different configurations. These feedback configurations include low-quality, single-dimensional, coarse-grained evaluation; high-quality, single-dimensional, coarse-grained evaluation; high-quality, multi-dimensional, coarse-grained evaluation; and high-quality, multi-dimensional, fine-grained evaluation. The dataset also includes human-generated reference summaries and key facts extracted from the documents using GPT4-turbo.
提供机构：
DISLab
