融合多视图图对比学习的中医脑卒中个性化处方推荐模型*
作者：赵紫一1，丁长松1,2
单位：1.湖南中医药大学信息科学与工程学院，湖南 长沙 410208； 2.湖南省中医药大数据分析实验室，湖南 长沙 410208
引用：引用：赵紫一，丁长松.融合多视图图对比学习的中医脑卒中个性化处方推荐模型[J].中医药导报,2025,31(4):231-237.
DOI：10.13862/j.cn43-1446/r.2025.04.044
PDF：
下载PDF
摘要：
目的：基于真实世界脑卒中临床数据，构建一种融合多视图图对比学习（MVCL）的个性化处方推荐模型。方法：根据脑卒中病例临床数据集建模多图结构，并融合中药属性及症状语义信息，以获得具有中医特征的节点嵌入向量。随后提出一种相似视图生成器用于视图增强，并在症状空间与中药空间分别开展局部对比学习，从而使模型能够更深入地挖掘数据中的配伍规律，提升处方推荐的准确性。结果：脑卒中数据集对比实验结果表明，相较于性能较优的SMRGAT，MVCL在Precision@5、Recall@5和F1-Score@5指标上分别提升了5.01%、5.43%和5.30%。公共数据集（PTM）对比实验结果表明，MVCL在Precision@5、Recall@5和F1-Score@5指标上分别提升了7.92%、9.94%和9.08%。结论：MVCL模型在推荐准确性和模型泛化能力方面均展现优势，为中医脑卒中的辅助诊疗决策提供了更可靠的支持及新的研究思路。
关键词：脑卒中；中医处方推荐；推荐模型；多视图；图对比学习；辅助诊疗
Abstract：
Objective: To construct a personalized prescription recommendation model that integrates multi-view graph contrastive learning (MVCL) based on real-world stroke clinical data. Methods: The stroke clinical dataset is modeled as a multi-graph structure, combining the properties of traditional Chinese medicine (TCM) and symptom semantic information to generate node embeddings with TCM characteristics. Subsequently, a similar view generator is proposed for view augmentation, and local contrastive learning is conducted separately in the symptom space and the TCM space to enable the model to uncover the compatibility patterns in the data better, thereby improving the accuracy of prescription recommendation. Results: The results of the comparative experiment on the stroke dataset show that compared with the relatively better performing SMRGAT, MVCL has improved by 5.01%, 5.43%, and 5.30% respectively in the Precision@5, Recall@5, and F1-Score@5 indicators. The results of the comparative experiment on the public dataset (PTM) show that MVCL has improved by 7.92%, 9.94%, and 9.08% respectively in the Precision@5, Recall@5, and F1-Score@5 indicators. Conclusion: The MVCL model shows advantages in both the accuracy of recommendation and the generalization ability of the model, providing more reliable support and new research ideas for the auxiliary diagnosis and treatment decision-making of stroke in traditional Chinese medicine.
Key words：stroke; prescription recommendation in traditional Chinese medicine; recommendation model; multi-view; graph contrastive learning; auxiliary diagnosis and treatment
发布时间：2025-12-20
点击量：24