目的 针对中医个性化处方推荐问题，研究自动化处方推荐任务，为中医临床辅助决策提供参考。方法 基于大语言预训练文本生成模型设计一种中医个性化处方推荐算法。将中医处方推荐任务转化为端到端(seq2seq)的文本生成任务，即将临床症状描述文本通过模型转化为处方文本，以实现处方推荐任务的需求，并利用基于大语言预训练的BART(Bidirectional and Auto-Regressive Transformers)模型的预训练参数来提升模型对通用语义信息的理解，通过对训练集处方内中药排序提升模型的处方推荐性能。结果 实验证明通过大语言预训练模型以及端到端的文本生成架构可有效提升模型的生成性能，同时对处方内中药依次排序可以获取更高准确率，并且通过中药的排列获取更多值得参考的有价值信息。中医个性化处方推荐模型在处方排序后分别在前5、10、15味生成的处方分别取得了 58。60、53。79和49。67的准确率。结论 中医个性化处方推荐模型取得了更优的处方推荐效果，表明其可为中医临床治疗疾病进行参考，达到辅助临床决策支持的效果。
Research on Personalized Prescription Recommendation of Traditional Chinese Medicine Based on Large Language Pre-Training Model
Objective Aiming at the problem of personalized prescription recommendation of TCM,the automatic prescription recommendation task is studied to provide reference for TCM clinical decision-making.Methods Based on the large language pre-trained text generation model,a personalized prescription recommendation algorithm of traditional Chinese medicine is de-signed.The TCM prescription recommendation task is transformed into an end-to-end(seq2seq)text generation task,that is,the clinical symptom description text is converted into prescription text through the model to realize the requirements of the pre-scription recommendation task,and the pre-training parameters of the Bidirectional and Auto-Regressive Transformers(BART)model based on large language pre-training are used to improve the model's understanding of general semantic information,and the prescription recommendation performance of the model is improved by ordering TCM medicines in the prescription of the train-ing set.Results Experiments show that the model generation performance can be effectively improved through the large language pre-training model and the end-to-end text generation architecture,and the sequential ordering of Chinese drug in the pre-scription can obtain more valuable information worthy of reference in the arrangement of Chinese drugs at the same time as higher accuracy.In this paper,the TCM personalized prescription recommendation model has achieved an accuracy rate of 58.60,53.79 and 49.67 in the top 5,10 and 15 herbs respectively after prescription ordering.Conclusion In this paper,the personalized pre-scription recommendation model of TCM has achieved better prescription recommendation effect,indicating that it can be used as a reference for the clinical treatment of diseases in TCM,and meet the needs of clinical auxiliary decision support.
prescription recommendationlarge language modelTCMtext generation
NETL
NSTL
万方数据
