基于BERT和领域词典融合的中文电子病历命名实体识别
Named Entity Recognition of Chinese Electronic Medical Record Based on BERT and Domain Dictionary
叶恩光 1张晓如 2张再跃 2丁腊春 3朱向南 3王译3
作者信息
- 1. 江苏科技大学 镇江 212003;镇江市第四人民医院 镇江 212001
- 2. 江苏科技大学 镇江 212003
- 3. 镇江市第四人民医院 镇江 212001
- 折叠
摘要
医疗数据挖掘的起始环节为CNER(中文电子病历命名实体识别),将相关实体(解剖部位、药品、影像检查等)识别出非结构化文本是其目标所在.基于CNER准确性提升需要,论文设计了BERT-BiLSTM-CRF模型融合领域词典技术,该技术能将上下文语义关系全面结合,一词多义问题同样可以迎刃而解,获取电子病历句子的长距离依赖.CNER采用BERT-BiLSTM-CRF模型融合领域词典技术时的F1 值已经被实验结果所证实,对知识图谱的构建、临床决策支持系统和病历质控系统等的研究有着重要意义.
Abstract
The beginning of medical data mining is CNER(named entity recognition of Chinese electronic medical record).The target of medical data mining is to recognize unstructured text from related entities(anatomical parts,drugs,image examina-tion,etc.).Based on the need of improving the accuracy of CNER,This paper designs the BERT-BiLSTM-CRF model fusion do-main dictionary technology,which can fully combine the context semantic relationship,solve the polysemy problem,and obtain the long-distance dependence of EMR sentences.When CNER uses the BERT-BiLSTM-CRF model to fuse the domain dictionary tech-nology,the value of F1 has been confirmed by the experimental results,which is of great significance to the construction knowledge graph,clinical decision support system and medical record quality control system.
关键词
中文电子病历/命名实体识别/BERT-BiLSTM-CRF/领域词典Key words
Chinese electronic medical record/named entity recognition/BERT-BiLSTM-CRF/domain dictionary引用本文复制引用
基金项目
江苏省重大科技示范项目(BE2018700)
国家自然科学基金(61371114)
出版年
2024
国家科技期刊平台
NSTL
万方数据