PortalPal-AI/Followup-Q

Followup-Q数据集是为了测试大型语言模型生成针对患者门户异步信息的后续问题的能力而构建的。每个信息都是由大型语言模型合成生成的，并与协作医院中患者的实际去标识化电子健康记录（EHR）配对。每套问题都是由临床专家编写的，这些问题可以包含一个或多个问题，以适应异步门户设置中的电子邮件式交流。临床专家在编写问题时，可以从EHR和消息中提取信息。该数据集包含训练集划分，共有250个示例，数据大小为561,028字节。数据集的语言为英文，主题标签包括医疗、医疗保健、问题和健康，属于小于1K的大小类别，并遵循MIT许可证。

The Followup-Q dataset was constructed to test the ability of large language models to generate sets of follow-up questions for asynchronous patient portal messages using both structured and unstructured data sources. Each message is synthetically generated by a large language model and paired with a real de-identified EHR from a patient at our collaborating hospital. Each set of questions is written by a clinical expert, and these follow-up questions are in sets because in an asynchronous portal setting, physicians respond with an email-like exchange that can contain one or more questions. Clinical experts are allowed to pull information from both the EHR and the Message when writing the questions. The dataset includes a training split with 250 examples, totaling 561,028 bytes in size. The language of the dataset is English, with tags including medical, healthcare, question, and health, categorized under size category n<1K, and licensed under the MIT license.