融合多特征嵌入的中文医疗命名实体识别模型MF-MNER  

MF-MNER:Chinese Medical Named Entity Recognition Model Fusing Multi-feature Embedding

作　　者：常远 季长伟 张春玲[1] 胡强[1] CHANG Yuan;JI Changwei;ZHANG Chunling;HU Qiang(School of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266001,China)

机构地区：[1]青岛科技大学信息科学技术学院,山东青岛266001

出　　处：《小型微型计算机系统》2024年第12期2915-2922,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61973180)资助;山东省重点研发计划(软科学)项目(2023RKY01009)资助。

摘　　要：医疗问诊数据中的命名实体识别不仅面临着实体交叉与边界模糊,而且问诊数据通常存在表述不准确、不规范和口语化等问题,已有医疗命名实体识别方法在问诊数据中适用效果较差.为此,提出一种适用于问诊数据的多特征嵌入中文医疗命名实体识别模型MF-MNER.该模型从字符、部首、词汇、边界和句法依赖等不同视角下获取字符的语义特征,并将融合后的语义特征经过扩张卷积神经网络进行卷积聚合,最后采用CRF模型进行序列解码.在医疗问诊数据集中开展的实验表明,多特征嵌入能明显提升命名实体的识别质量,MF-MNER相对于其他方法能够更适用于问诊数据中的医疗命名实体识别.此外,在公开的电子病例集中的实验表明,MF-MNER的高性能医疗命名实体识别具有普适性.Medical named entity recognition in consultation data not only faces the problems of entity intersection and fuzzy boundaries,but also has the problems of inaccurate representation,non-standard and colloquialism in consultation data.Existing medical named entity recognition methods have poor application effects in consultation data.To solve this problem,this paper proposed a multi-feature embedded Chinese medical named entity recognition model MF-MNER suitable for consultation data.This model obtained the semantic features of characters from different perspectives such as characters,radicals,words,boundaries and syntactic dependencies,and then aggregated the fused semantic features through the dilated convolutional neural network.Finally,it used the CRF model to sequence decoding.Experiments on the online medical consultation data set show that multi-feature embedding can significantly improve the recognition quality of named entities,and MF-MNER is more suitable for medical named entity recognition in consultation data than other methods.In addition,the experiments in the public electronic medical record set show that the high performance of MF-MNER for medical named entity recognition is universal.

关 键 词：命名实体 问诊数据 多特征嵌入 扩张卷积神经网络 CRF模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]