基于ERNIE-BiGRU-Attention-CRF的电子病历命名实体识别方法  

Named Entity Recognition Method of Electronic Medical Records Based on ERNIE-BiGRU-Attention-CRF

作　　者：王正芳 张军亮[1,2] 李小倩 于月 陈慧媜 WANG Zhengfang;ZHANG Junliang;LI Xiaoqian;YU Yue;CHEN Huizheng(School of Health Management,Xinxiang Medical University,Xinxiang 453003,China;Center for Health Information Resources,Xinxiang Medical University,Xinxiang 453003,China)

机构地区：[1]新乡医学院卫生健康管理学院,新乡453003 [2]新乡医学院卫生信息资源研究中心,新乡453003

出　　处：《医学信息学杂志》2024年第5期76-82,100,共8页Journal of Medical Informatics

基　　金：国家社会科学基金项目(项目编号:21BTQ051);新乡医学院研究生科研创新支持计划一般项目(项目编号:YJSCX202285Y);河南省高校哲学社会科学创新人才支持计划(项目编号:2023-CXRC-13)。

摘　　要：目的/意义改善中文电子病历命名实体识别模型的性能,更好地开展医疗信息的组织和挖掘。方法/过程构建ERNIE-BiGRU-Attention-CRF中文电子病历命名实体识别模型,首先采用ERNIE1.0预训练模型生成具有语义特征的词向量,然后利用BiGRU捕获全局语义特征与语法结构特征,通过Attention机制进一步增强语义特征的捕获,最后连接CRF解码层输出全局概率最大的标签序列。结果/结论在公开的医疗文本数据集CCKS2017开展对比实验、消融实验,利用生成的模型进行实例分析,取得较好的识别效果。Purpose/Significance To improve the performance of named entity recognition(NER)model of Chinese electronic medical records(EMR)for better organization and mining of medical information.Method/Process The ERNIE-BiGRU-Attention-CRF NER model of Chinese EMR is constructed.Firstly,the ERNIE1.0 pre-training model is used to generate word vectors with semantic features,and then BiGRU is utilized to capture the global semantic features and grammatical structural features,which are fed into the Attention mechanism to further enhance the capture of the semantic features,and finally,the CRF decoding layer is connected to output the label sequences with the maximum global probability.Result/Conclusion Comparison experiments and ablation experiments are carried out on the publicly available medical text dataset CCKS2017,and examples analysis is conducted using the generated model.The model proposed in this paper achieves better recognition results.

关 键 词：命名实体识别 ERNIE 双向门控循环神经网络 注意力机制 条件随机场 

分 类 号：R-058[医药卫生]