Research on the Construction of TCM Diagnosis Model based on Large Language Model

To solve many problems in TCM syndrome differentiation, including the lack of public data and quality differences, the problem of singleness and universality of models, and the lack of interpretability of models, a solution based on large language model ChatGLM3 combined with LoRA fine-tuning technology was proposed. The open source TCM-SD TCM syndrome differentiation data set was adopted, and after data filtering and integration optimization, 1027 TCM syndrome differentiation definition data sets, 41180 consultation training data and 5485 testing and verification data were obtained, so that the model could deeply learn the specialized knowledge of TCM syndromes and the actual consultation records of TCM syndrome differentiation. The experimental results show that the evaluation indexes of two different trainings using LoRA fine-tuning technology are significantly improved by about 20%.