中文电子病历命名实体识别方法研究  

Study on Named Entity Recognition of Chinese Electronic Medical Records

作　　者：陈婕卿 竹志超 张锋 曾可 姜会珍 程振宁 CHEN Jieqing;ZHU Zhichao;ZHANG Feng;ZENG Ke;JIANG Huizhen;CHENG Zhenning(Peking Union Medical College Hospital,Chinese Academy of Medical Sciences&Peking Union Medical College,Beijing 100730,China;Beijing University of Technology,Beijing 100124,China;Analyzefocus Information Consultant Ltd.,Beijing 100005,China)

机构地区：[1]中国医学科学院北京协和医院信息中心,北京100730 [2]北京工业大学信息学部,北京100124 [3]北京安妮福克斯信息咨询有限公司,北京100005

出　　处：《医学信息学杂志》2024年第4期78-84,共7页Journal of Medical Informatics

基　　金：科技创新2030——“新一代人工智能”重大项目(项目编号:2020AAA0104900)。

摘　　要：目的/意义探索基于中文电子病历的命名实体识别方法在构建医学知识图谱和相关应用推广方面的技术可行性。方法/过程采用真实医疗电子病历数据对词嵌入表示模型进行精化,构建医学术语专有嵌入表示,并利用卷积神经网络等多模型提取局部语义特征,实现基于堆叠注意网络的中文医疗命名实体识别。结果/结论堆叠注意网络模型F1值达到91.5%,较其他模型具备更强的医疗命名实体识别性能。进一步解决中文医疗命名实体识别难点,在实现全局语义特征全面深入提取的同时降低时间成本。Purpose/Significance To explore the technical feasibility of named entity recognition(NER)method based on Chinese electronic medical records(EMR)in the construction of medical knowledge graph and related application promotion.Method/Process The word embedding representation model is refined by using real EMR data,and the proprietary embedding representation of medical terms is constructed.Moreover,multiple models such as convolutional neural network(CNN)are used to extract local semantic features to realize the recognition of Chinese medical named entities based on stacked attention network(SAN).Result/Conclusion The F 1 value of SAN model reaches 91.5%,which has stronger performance of medical NER than other models,so as to further solve the difficulty of Chinese medical NER,achieve comprehensive and in-depth extraction of global semantic features,and reduce the time cost.

关 键 词：电子病历 命名实体识别 堆叠注意网络 

分 类 号：R-058[医药卫生]