Meta Evaluation Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/amazon-science/RAGChecker
下载链接
链接失效反馈官方服务：
资源简介：
该数据集由280个实例构成，基于8个基准RAG系统的反馈进行成对的人类偏好标注。每个实例都由两位标注者根据正确性、完整性和整体评估进行标注。规模上，该数据集包含280个实例，其任务是对RAG系统响应进行人类偏好标注。
This dataset consists of 280 instances, with pairwise human preference annotations conducted based on feedback from 8 baseline RAG systems. Each instance is annotated by two annotators according to correctness, completeness and overall evaluation. In terms of scale, this dataset contains 280 instances, and its task is to perform human preference annotation on the responses of RAG systems.
提供机构：
Open-source community
