融合BERT和GCN的中医问诊辅助诊断算法研究  

Combining BERT and GCN for Supplementary Diagnosis of Traditional Chinese Medicine

作　　者：何圆姣 刘国华[1] He Yuanjiao;Liu Guohua(Engineering Research Center of Thin Film Optoelectronics Technology of Ministry of Education,Key Laboratory of Photoelectric Sensors and Sensor Network Technology,School of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津市光电传感器与传感网络技术重点实验室,薄膜光电子技术教育部工程研究中心,天津300350

出　　处：《南开大学学报（自然科学版）》2024年第2期70-78,共9页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：中央高校基本科研业务费专项资金资助。

摘　　要：针对中医病例样本数据属性特征庞杂,结构特征复杂度高的特点,构建高精度中医诊断算法模型BT-GCN.首先利用中医数字化古籍资料训练BERT自然语言处理模型生成中医术语词向量,与主次症、严重程度等指标共同作为病例输入特征;其次,设计图卷积神经网络(Graph Convolution Networks,GCN),二维卷积层作为学习层完成模型搭建.结果表明:BT-GCN模型在测试样本上的准确率高于其他模型,达到了97.6%.因此,BT-GCN在提取中医病例样本的有效特征和疾病分类诊断方面具有比较明显的应用优势.A high-precision algorithm Model BT-GCN for TCM diagnosis is constructed according to the characteristics of complex attribute and complex structure of TCM case data.Firstly,the natural language processing(NLP)model of Bert(bidirectional encoder representatives from Transformers)was used to generate the Chinese medical term vector,which was used as the case input feature together with the index of primary and secondary symptoms,severity,etc.,design convolutional neural network GCN(graph convolution networks),2D convolution layer is modeled as learning layer.The results show that the accuracy of BT-GCN is 97.6%higher than that of other models.Therefore,BT-GCN has obvious application advantages in extracting effective features of TCM cases and disease classification and diagnosis.

关 键 词：表示学习 自然语言处理 图卷积算法 中医诊断 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]