Leveraging Large Language Models for Automated Knowledge Acquisition in Personal Health Status Evaluation

Healthcare issues have always been regarded as an important application area of ​​artificial intelligence research. When an individual's physical abnormality occurs without sufficient medical knowledge, it often becomes more serious because it is not treated immediately and correctly, especially in remote areas where medical resources are scarce. Motivated by this, in this study we developed a personal healthcare assistant system which integrates Large Language Models (LLMs) to assess users' health status and provide appropriate assistance and advice. Users interact with the system by describing the discomfort they are experiencing. Based on these descriptions, the system matches symptoms with diseases using a similarity comparison from a hospital dataset. It selects the 20 most relevant conditions as the basis for the model's few-shot learning and then employs Instruction Learning to tailor the model for this assessment task. This system lets users understand the possible diseases corresponding to their symptoms preliminarily and follow the system's instructions for timely and appropriate treatment. Additionally, this study evaluates the performance of various LLMs, such as GPT-4, Llama3, and TAIDE models. We assess the models based on thousands of interactions between the system and users, comparing the applicability and differences of different models. Experimental results demonstrate that the proposed system can achieve the basic functions of personal health monitoring and assessment of health status.