A Study on Multi-Label Classification Methods for Traditional Chinese Medicine Literature Based on Sentence Embedding Enhanced by Graph Neural Networks
- VernacularTitle: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
- Author:
Jingyao CHEN
1
;
Jinghua LI
1
;
Tong YU
1
Author Information
- Publication Type:Journal Article
- Keywords: Multi-label Classification; Imbalanced Sampling; TCM Heterogeneous Network; Graph Neural Network
- From: World Science and Technology-Modernization of Traditional Chinese Medicine 2025;27(2):420-430
- CountryChina
- Language:Chinese
- Abstract: 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.
