DRAGONBall
收藏资源简介:
DRAGONBall数据集由清华大学等机构创建,是一个多领域、多语言的评估数据集,专门用于测试RAG模型在金融、法律和医疗领域的知识使用能力。该数据集包含6711个问题,涵盖20个金融领域、10个法律领域和19个医疗类别,文本内容包括中文和英文。数据集的创建过程涉及从种子文档中总结模式,生成多样化的文档,并根据文档和配置构建问题-答案对,旨在通过全面的评估框架提高RAG模型在特定领域的表现。
The DRAGONBall dataset, developed by Tsinghua University and other institutions, is a multi-domain, multilingual evaluation dataset specifically designed to test the knowledge utilization capabilities of Retrieval-Augmented Generation (RAG) models in financial, legal, and medical fields. It contains 6,711 questions covering 20 financial domains, 10 legal domains, and 19 medical categories, with text content available in both Chinese and English. The dataset creation process involves summarizing patterns from seed documents, generating diverse documents, and constructing question-answer pairs based on the documents and predefined configurations, aiming to improve the performance of RAG models in specific domains through a comprehensive evaluation framework.

- 1RAGEval: Scenario Specific RAG Evaluation Dataset Generation Framework清华大学, 北京师范大学, 中国科学院大学, 东北大学 · 2024年



