基于多神经网络协作的电子病历命名实体识别方法  

NAMED ENTITY RECOGNITION METHOD OF ELECTRONIC MEDICAL RECORD BASED ON MULTI NEURAL NETWORK COOPERATION

作　　者：张运中 纪斌 余杰[2] 刘慧君 Zhang Yunzhong;Ji Bin;Yu Jie;Liu Huijun(Hunan Electronic Port Service Center,Changsha 410001,Hunan,China;School of Computer,National University of Defense Technology,Changsha 410073,Hunan,China;Institute of Computer Application,China Academy of Engineering Physics,Mianyang 621999,Sichuan,China)

机构地区：[1]湖南省电子口岸服务中心,湖南长沙410001 [2]国防科技大学计算机学院,湖南长沙410073 [3]中国工程物理研究院计算机应用研究所,四川绵阳621999

出　　处：《计算机应用与软件》2021年第2期179-184,共6页Computer Applications and Software

基　　金：装备预研领域基金项目(31511070404,61400040201)。

摘　　要：随着电子病历在医疗领域的推广应用,越来越多的研究者关注如何高效地从电子病历中抽取高价值科研信息。CHIP2018将中文电子病历临床医疗命名实体识别作为评测任务,即从中文电子病历中抽取三种恶性肿瘤相关的实体。结合三种实体的特点和实体间的依赖关系,提出基于多神经网络协作的复杂医疗命名实体识别方法,并实现了句子级别的模型迁移,解决了训练数据集数量和质量问题,最终获得了该评测任务的第二名。此外,该方法的改进方法取得了CCKS2019评测任务一的第一名,印证了其有效性和泛化能力。With the application of electronic medical records in medical field,more and more researcher are paying attention to how to efficiently extract high-value scientific research information from electronic medical records.The CHIP2018 takes Chinese electronic medical record clinical medical named entity recognition as an open challenge,specifically,extracts three malignant tumor-related entities from Chinese electronic medical records.Combining the characteristics of three entities and entities dependencies,we propose a complex medical named entity recognition approach based on multi neural network cooperation.It realized sentence-level model transfer application to solve the quantity and quality problem of data set.And our approach got the second place in the evaluation task.In addition,our approach won the champion of the first evaluation task released by CCKS2019,which further validate the effectiveness and generalization ability.

关 键 词：神经网络 BiLSTM-CRF 中文电子病历 命名实体识别 模型迁移 泛化 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]