特征加权与子网络动态组合的中医辨证模型  

Traditional Chinese medicine syndrome differentiation model based on feature weighting and sub-network dynamic combination

作　　者：张素华 叶青[1] 程春雷[1] 邹静 方桦 ZHANG Su-hua;YE Qing;CHENG Chun-lei;ZOU Jing;FANG Hua(School of Computer Science,Jiangxi University of Chinese Medicine,Nanchang 330004,China;Educational Technology and Teaching Resource Center,Nanchang University,Nanchang 330004,China)

机构地区：[1]江西中医药大学计算机学院,南昌330004 [2]南昌大学教育技术与教学资源中心,南昌330004

出　　处：《信息技术》2023年第11期56-61,共6页Information Technology

基　　金：国家重点研发计划课题(2019YFC1712301)。

摘　　要：临床中的中医电子病历数据描述存在噪音,如何解决不同的噪声水平和四诊信息之间的冲突是一个挑战。对此,提出了基于特征加权与子网络动态组合的中医辨证模型。模型首先基于TF-IDF算法对字段中文本计算权重,然后将脉诊、问诊、望诊(其中望诊中的舌诊在病历中为单独字段)、主诉字段输入相应的子网络中,计算子网络的输出与TF-IDF权重的权值注意力,接下来将各个子网络在每个病历动态变化的一组权重向量进行动态加权组合。在中医电子病历数据集上的实验结果表明,模型辨证准确率较高,具有更显著的中医辨证效果。There is noise in the description of traditional Chinese medicine(TCM)electronic medical record data in clinical practice,and how to solve different noise levels and conflicts between the four diagnostic information is a challenge.In this regard,this paper proposes a TCM syndrome differentiation model based on feature weighting and dynamic combination of sub-networks.The model first calculates the weight of the text in the field based on the TF-IDF algorithm,and then inputs the fields of pulse diagnosis,inquiry,inspection(the tongue diagnosis in the inspection is a separate field in the medical record),and chief complaint into the corresponding sub-network,and the output of the sub-network is weighted with the TF-IDF weight.The attention calculation is followed by a dynamic weighted combination of a set of weight vectors that each sub-network dynamically changes in each medical record.The experiment results on the TCM electronic medical record dataset show that the proposed model improves the accuracy of syndrome differentiation and has a more significant effect of TCM syndrome differentiation.

关 键 词：中医辨证 特征加权 子网络动态组合 中医电子病历 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]