基于图神经网络增强句嵌入的中医文献多标签分类方法研究  

A Study on Multi-Label Classification Methods for Traditional Chinese Medicine Literature Based on Sentence Embedding Enhanced by Graph Neural Networks

作　　者：陈靖耀 李敬华[1] 于彤[1] CHEN Jingyao;LI Jinghua;YU Tong(Institute of Information on Traditional Chinese Medicine,Beijing 100700,China)

机构地区：[1]中国中医科学院中医药信息研究所,北京100700

出　　处：《世界科学技术-中医药现代化》2025年第2期420-430,共11页Modernization of Traditional Chinese Medicine and Materia Medica-World Science and Technology

基　　金：中国中医科学院基本科研业务费自主选题项目(ZZ150313):中医药知识图谱自动构建共性技术研究与平台开发,负责人:于彤;中国中医科学院科技创新工程重大攻关项目(CI2021A05308):基于海量文献的中医脾胃病知识图谱自动构建与知识服务研究,负责人:于彤;中国中医科学院科技创新工程中医药信息学创新团队(CI2021B002):大规模中医药知识图谱的构建、融合与有效方药发现研究,任务负责人:李敬华。

摘　　要：目的提出一种使用图神经网络增强句嵌入的中医文献多标签分类方法,能够有效获取相似文章之间的关联,通过与文本语义信息的相互融合,提高分类性能。方法获取论文句嵌入数据,建立中医文献异构网络,通过图神经网络GraphSAGE模型学习论文在异构网络上的表征信息和自身的句嵌入信息,将得到的特征向量输入模型进行多标签分类。结果在中医文献数据集下,基于图神经网络的模型的多标签分类精确率和F1值,达到了0.83与0.72,优于主流基线模型。结论本文提出的方法在中医文献多标签分类任务中具有有效性。Objective We propose a method for multi-label classification of traditional Chinese medicine(TCM)literature using graph neural networks to enhance sentence embeddings.This approach can effectively capture the relationships between similar articles.By integrating with the semantic information of the text,it improves classification performance.Methods Sentence embedding data of papers are obtained,and a heterogeneous network of traditional Chinese medicine literature is established.The representation information of papers on the heterogeneous network and their own sentence embedding information are learned through the GraphSAGE model of graph neural networks.The feature vectors obtained are then input into the model for multi-label classification.Results In a TCM literature dataset,the multi-label classification model based on graph neural networks achieved precision and F1 scores of 0.83 and 0.72,respectively,outperforming mainstream baseline models.Conclusion The effectiveness of the proposed method in the multi-label classification task for TCM journals.

关 键 词：多标签分类 样本不均衡 中医文献异构网络 图神经网络 

分 类 号：R-058[医药卫生]