FaithfulnessQAC, UniqueQAC
收藏资源简介:
ASTRID数据集由Ufonia Limited和约克大学的研究团队创建,旨在评估基于检索增强生成(RAG)的临床问答系统的性能。数据集包含来自真实患者的临床问题,涵盖了白内障手术后随访中的常见问题,并补充了急诊、临床和非临床领域的问题。FaithfulnessQAC数据集包含238条问题-答案-上下文三元组,UniqueQAC数据集包含132条三元组。数据集的创建过程包括从临床对话中提取问题,并由临床医生选择相关问题以增强数据集的多样性和覆盖范围。该数据集的应用领域主要集中在临床问答系统的自动评估,旨在解决现有评估指标在临床和对话场景中的不足,确保生成的回答在临床上是准确且有用的。
The ASTRID dataset was created by a research team from Ufonia Limited and the University of York, aiming to evaluate the performance of clinical question-answering systems based on Retrieval-Augmented Generation (RAG). The dataset contains clinical questions from real patients, covering common issues during post-operative follow-up after cataract surgery, supplemented by questions from emergency, clinical, and non-clinical fields. The FaithfulnessQAC dataset includes 238 question-answer-context triples, while the UniqueQAC dataset contains 132 triples. The dataset creation process involves extracting questions from clinical dialogues and having clinicians select relevant questions to enhance the diversity and scope of the dataset. The application domain of this dataset is primarily focused on the automatic evaluation of clinical question-answering systems, aiming to address the deficiencies of existing evaluation metrics in clinical and conversational scenarios, ensuring that the generated answers are accurate and useful in a clinical setting.




