HAI-Coaching

HAI-Coaching是首个专家标注的营养咨询数据集，由意大利卡利亚里大学创建。该数据集包含约2.4K来自众包工作者的饮食挣扎案例和约97K由ChatGPT生成的相关支持文本。数据集的创建过程涉及众包收集和聚类饮食相关问题，随后与专家合作引导ChatGPT生成支持文本，并由专家评估文本的安全性。HAI-Coaching旨在解决营养咨询领域数据稀缺的问题，并探索在敏感领域如健康领域中安全使用语言模型的方法。

HAI-Coaching is the first expert-annotated nutritional counseling dataset developed by the University of Cagliari, Italy. This dataset contains approximately 2.4K cases of dietary struggles collected from crowdsourced workers, as well as roughly 97K relevant supportive texts generated by ChatGPT. The dataset creation process involved crowdsourcing and clustering diet-related questions, followed by collaborating with domain experts to guide ChatGPT in generating supportive texts, which were subsequently evaluated for safety. HAI-Coaching aims to address the data scarcity issue in the nutritional counseling field, and explore methods for the safe utilization of language models in sensitive domains such as healthcare.