太原理工大学学报2025，Vol.56Issue(5)：866-874,9.DOI:10.16355/j.tyut.1007-9432.20240178
基于TCMTF的中草药推荐方法研究
Recommendation Method of Chinese Herbal Medicine Based on TCMTF
摘要
Abstract
[Purposes]Taking the task of Chinese herbal medicine recommendation as the starting point,a Chinese herbal medicine recommendation model based on an improved Transformer is pro-posed to address the problems of existing Chinese herbal medicine recommendation models ignoring traditional Chinese medicine related theoretical knowledge,resulting in poor effectiveness and devia-tion from actual recommended prescriptions.[Methods]Convolutional neural networks were added to the symptom sequence information mining module,the multi branch structure of the model was ex-panded,and the trained symptom text vector was used as the embedding vector to mine relevant infor-mation to improve the deep level feature combination ability and recommendation quality of the model.A multi feature fusion attention mechanism was proposed to reduce the feature dimensions that are overly focused during the attention accumulation process,so that the accumulation results can focus on areas that are not being paid attention to and reduce the probability of recommending Chinese herbal medicine duplicates.An entropy smoothing loss was also proposed to further improve recom-mendation results by reducing the impact of strict order on the results.The model was tested on a pub-lic clinical Chinese medicine prescription dataset and a private dataset of a collaborating hospital.[Re-sults]The experimental results show that compared with other Chinese herbal recommendation mod-els such as Herb Know and TCM Translator,the proposed model can effectively improve the quality of recommended Chinese herbal medicine and improve the problem of repetition in the process.Com-pared with the best benchmark model,the proposed model improved the precision,recall,and F1 by 7%-9%,5%-6%,and 7%-8%,respectively.In addition,the ablation experiment also demon-strates the effectiveness between various modules.关键词
中草药推荐/Transformer/熵平滑损失/多特征融合注意力机制Key words
recommendation of Chinese herbal medicine/Transformer/entropy smoothing loss/multi-feature fusion attention mechanism分类
信息技术与安全科学引用本文复制引用
侯校,杨丰豪,赵紫娟,朱晓军,强彦..基于TCMTF的中草药推荐方法研究[J].太原理工大学学报,2025,56(5):866-874,9.基金项目
国家自然科学基金(U21A20469,61872261,61972274) （U21A20469,61872261,61972274）
