融合注意力机制的蒙医药命名实体识别

Named Entity Recognition of Mongolian Medicine with Integrated Attention Mechanism

刘忠博 ¹杨一帆 ¹白青海 ¹周玉新 ¹刁宇峰 ¹张军¹

扫码查看

作者信息

1. 内蒙古民族大学计算机科学与技术学院,内蒙古通辽 028043
折叠

摘要

目前蒙医药文本分布较为分散,缺乏系统化整理,构建蒙医药知识图谱成为解决这一问题的关键途径,其中命名实体识别(NER)技术发挥着关键作用.提出了一种基于BERT-BiGRU-CRF与注意力机制的NER模型,旨在解决蒙医药文本中实体识别问题,数据来源包括公开发布的蒙医药文本以及蒙医的著作,并进行校正和完善.实验结果表明,所提出的模型在蒙医药命名实体识别任务中的F1值达到87.33%.在F1值上,相比于BiLSTM-CRF、BERT-BiLSTM-CRF、BERT-BiGRU-CRF模型分别提升了4.97%、1.82%和1.77%,不仅提升了蒙医药领域中NER的应用效果,还为蒙医药知识图谱的构建和文化传承提供了重要的技术支持.

Abstract

At present,Mongolian medicine texts are scattered and lack systematic organization.The key solu-tion to solve this problem is to construct Mongolian medicine knowledge map,in which Named Entity Recognition(NER)technology plays a key role.In this paper,a NER model based on BERT-BiGRU-CRF and attention mecha-nism is proposed to solve the entity recognition issue in Mongolian medicine texts.The data sources include the published Mongolian medical texts and Mongolian medical works which have been corrected and improved.Experi-mental results show that the proposed model achieves the F1 value of 87.33%in the Mongolian medicine Named Entity Recognition task.The F1 score represents an improvement of 4.97%,1.82%,and 1.77%compared to the BiL-STM-CRF,BERT-BiLSTM-CRF,and BERT-BiGRU-CRF models,respectively.This research not only enhances the application of NER in the field of Mongolian medicine but also provides important technical support for the con-struction of Mongolian medicine knowledge graphs and cultural heritage preservation.

关键词

蒙医药知识图谱/命名实体识别/注意力机制/BERT

Key words

Mongolian medicine knowledge graph/Named Entity Recognition/attention mechanism/BERT

引用本文复制引用

出版年

2025

内蒙古民族大学学报(自然科学版)

内蒙古民族大学

内蒙古民族大学学报(自然科学版)

影响因子：0.444

ISSN：1671-0185

段落导航