全文下载:
202605018.pdf
文章编号:1672-6987(2026)05-0152-07;DOI:10. 16351/j. 1672-6987. 2026. 05. 018
何洪吉,胡乃平* (青岛科技大学 信息科学技术学院,山东 青岛 266061)
摘 要:电子病历命名实体识别(EMR)是医疗信息抽取的重要基础。针对现有中文电子病 历实体识别模型存在特征提取能力不足、词汇边界模糊等问题。本文提出了一种基于全局指 针(global pointer,GP)和特征融合的命名实体识别模型(MCGP),模型将通过 MacBERT 提取 的字向量和用 Word2Vec 提取的词向量进行融合,获得词段级和句子级的深层文本语义信息, 然后通过旋转编码层添加相对位置信息,最后使用全局指针进行解码,将待识别实体的首部 和 尾 部 作 为 一 个 统 一 的 整 体 进 行 判 别 。 实 验 结 果 表 明 ,与 现 有 主 流 模 型 相 比 ,模 型 在 CCKS2019 数据集上的 F1 值达 86. 71%,明显优于其它命名实体识别模型。
关键词:命名实体识别;中文电子病历;特征嵌入;全局指针
中图分类号:TP 391. 1 文献标志码:A
引用格式:何洪吉,胡乃平 . 基于全局指针与特征融合的中文电子病历命名实体识别[J]. 青 岛科技大学学报(自然科学版),2026,47(5):152-158.
HE Hongji, HU Naiping. Named entity recognition for Chinese electronic medical records based on feature embedding and global pointer[J]. Journal of Qingdao University of Science and Technology(Natural Science Edition),2026,47(5):152-158.
Named Entity Recognition for Chinese Electronic Medical Records Based on Feature Embedding and Global Pointer
HE Hongji,HU Naiping (College of Information Science and Technology, Qingdao University of Science and Technology, Qingdao 266061, China)
Abstract: Electronic medical record named entity recognition (EMR) is an important foundation for medical information extraction. For the existing Chinese electronic medical record entity recognition model has the problems of insufficient feature extraction ability and blurred vocabulary boundary. In this paper, we propose a named entity recognition model (MCGP) based on GlobalPointer (GP) and feature fusion. The model fuses word vectors extracted by MacBERT and word vectors extracted with Word2Vec to obtain deep text semantic information at word segment level and sentence level, then adds relative position information by Rotary Positional Embedding layer, and finally uses GlobalPointer for decoding to discriminate the first and last parts of the entity to be recognized as a unified whole. The experimental results show that the model achieves an F1 value of 86. 71% on the CCKS2019 dataset compared with existing mainstream models, which is significantly better than other named entity recognition models.
Key words:named entity recognition; chinese electronic record; feature embedding; global pointer
收稿日期:2025-10-25
基金项目:国家自然科学基金项目(61104004).
作者简介:何洪吉(1998—),男,硕士研究生 . * 通信联系人 .