现代情报2025，Vol.45Issue(5)：24-33,10.DOI:10.3969/j.issn.1008-0821.2025.05.004
面向中文病历的实体关系抽取模型研究
Research on Entity Relationship Extraction Model for Chinese Medical Record
摘要
Abstract
[Purpose/Significance]Relational extraction is a core component of medical record processing,which is crucial for improving the accuracy and efficiency of electronic medical record processing.In order to solve the problems of entity redundancy,entity word nesting and entity overlapping in the relational extraction of Chinese electronic medical re-cords,and to improve the efficiency of medical information extraction,a novel relational extraction model for Chinese medi-cal records is proposed.[Method/Process]The relationship extraction task was decomposed into three parts:relationship prioritization decoder,global entity extraction and subject-object alignment.Firstly,the relationship was predicted and fil-tered by the decoder,and the entity extraction was restricted based on the predicted relationships.Secondly,the relation-ship-specific attention mechanism and the global pointer network were adopted to effectively deal with the problem of infor-mation overlapping and subject/object nesting.Finally,the entity correspondence matrix was introduced to align the sub-ject,object,and their relationships into a ternary group.[Result/Conclusion]Comprehensive experiments are conducted on CMeIE Chinese medical record dataset and DiaKG real diabetes Chinese dataset respectively and compared with six mod-els,and it is found that the F1 values of this paper's model on the datasets CMeIE and DiaKG are improved by 6.6%and 5.8%compared with the mainstream model CasRel,respectively.The results show that the model in this paper can effec-tively solve the problems of entity nesting and entity overlapping caused by the complexity of Chinese medical records,which is valuable for medical information extraction and data processing processes.关键词
关系抽取/中文病例/电子病例/实体嵌套/实体重叠/注意力机制/全局指针/糖尿病Key words
relation extraction/Chinese case studies/electronic medical records/entity nesting/entity overlap/attention mechanism/global pointers/diabetes分类
社会科学引用本文复制引用
单涛,许鑫,王园梦,王宇翱,景慎旗,叶继元,郭永安..面向中文病历的实体关系抽取模型研究[J].现代情报,2025,45(5):24-33,10.基金项目
国家重点研发计划"主动健康服务数治化技术区域综合应用示范"(项目编号:2023YFC3605800) （项目编号:2023YFC3605800）
国家社会科学基金重大课题"新时代我国文献信息资源保障体系重构研究"(项目编号:19ZDA346) （项目编号:19ZDA346）
江苏省前沿引领技术基础研究专项"人机物深度融合高可信网构软件技术、理论与方法"(项目编号:BK20202001). （项目编号:BK20202001）
