图书情报工作

图书情报工作 >

2025 , Vol. 69 >Issue 5: 107 - 116

DOI: https://doi.org/10.13266/j.issn.0252-3116.2025.05.010

研究论文

面向中文医学命名实体识别的判别式与生成式语言模型比较研究

  • 刘伟 ,
  • 薛航 ,
  • 张晗
展开
  • 中国医科大学健康管理学院 沈阳 110122
刘伟，副教授，硕士；薛航，硕士研究生；张晗，教授，博士，硕士生导师，通信作者，E-mail:zhanghan@cmu.edu.cn。

收稿日期: 2024-07-03

  修回日期: 2024-09-16

  网络出版日期: 2025-03-07

基金资助

本文系辽宁省教育厅基本科研项目“基于深度学习的线上用户慢性病健康教育问答模型研究”（项目编号：LJKR0275）研究成果之一。

收起

A Comparative Study of Discriminative Language Model and Generative Language Model for Chinese Medical Named Entity Recognition

  • Liu Wei ,
  • Xue Hang ,
  • Zhang Han
Expand
  • School of Health Management, China Medical University, Shenyang 110122
Liu Wei,associate professor,master;Xue Hang,master candidate;Zhang Han,professor,PhD,master supervisor,corresponding author,E-mail:zhanghan@cmu.edu.cn.

Received date: 2024-07-03

  Revised date: 2024-09-16

  Online published: 2025-03-07

Supported by

This work is supported by the 2021 Scientific Research Project of Educational Department in Liaoning Province titled “Deep Learning-based Question Answering Model for Online Education of Chronic Disease Users” (Grant No. LJKR0275).

Fold

摘要

[目的/意义] 比较以BERT为代表的判别式语言模型与以ChatGPT为代表的生成式语言模型完成中文医学文本命名实体识别任务的效果，以期为生物医学领域的命名实体识别提供参考。[方法/过程] 设计6种BERT拼接不同类型深度学习网络的抽取模型，同时针对ChatGPT设计融合不同维度元素的提示语，采用精确匹配和宽松匹配两种方式对中文医学实体的抽取效果进行评价。[结果/结论] 在判别式语言模型中，BERT- BiLSTM-CRF模型表现最优，其精确匹配与宽松匹配F1值分别达到85.71%和89.49%；在生成式语言模型中，结合提示语框架的GPT_baseline+A+three-shot模型取得最佳效果，其精确匹配与宽松匹配F1值分别为78.14%和84.73%。本研究场景下，BERT系列模型取得相对好的效果。精心设计的ChatGPT提示语框架能显著增强模型对命名实体识别任务的理解，有望成为未来自然语言处理领域的新突破点。

关键词： 命名实体识别; 深度学习网络; 生成式语言模型; 判别式语言模型

本文引用格式

刘伟 , 薛航 , 张晗 . 面向中文医学命名实体识别的判别式与生成式语言模型比较研究[J]. 图书情报工作, 2025 , 69(5) : 107 -116 . DOI: 10.13266/j.issn.0252-3116.2025.05.010

Abstract

[Purpose/Significance] To provide reference for named entity recognition in biomedical field, this article compares the effect of the discriminative language model represented by BERT and the generative language model represented by ChatGPT in Chinese medical texts named entity recognition tasks. [Method/Process] This article designed six models of BERT splicing different types of deep learning networks. Meanwhile, prompts with different dimensional elements were designed for ChatGPT, and the extracting results were evaluated by exact matching and loose matching respectively. [Result/Conclusion] Among the discriminative language models, BERT-BiLSTM-CRF model has the best performance, with F1 values of exact match and loose match reaching 85.71% and 89.49% respectively. In generative language models, GPT_baseline+A+three-shot model that combined with prompt language framework achieves the best results, and its exact match and loose match F1 values are 78.14% and 84.73% respectively. In this study, BERT series model achieves relatively best results. The well-designed prompt framework for ChatGPT could significantly enhance the model’s understanding of named entity recognition tasks, and is expected to be a new breakthrough in natural language processing tasks in the future.

Key words： named entity recognition; deep learning network; generative language model; discriminative language model

参考文献

[1] LEE J, YOON W, KIM S, et al. BioBERT: a pre-trained biomedical language representation model for biomedical text mining[J]. Bioinformatics, 2020, 36(4): 1234-1240.
[2] GU Y, TINN R, CHENG H, et al. Domain-specific language model pretraining for biomedical natural language processing[J]. ACM transactions on computing for healthcare, 2021, 3(1): 1-23.
[3] 徐凯, 王崎, 李振彰, 等. 基于结合多头注意力机制BiGRU网络的生物医学命名实体识别[J]. 计算机应用与软件, 2020, 37(5): 151-155. (XU K, WANG Q, LI Z Z, et al. Biomedical named entity recognition based on BiGRU network with multi-head attention mechanism[J]. Computer applications and software, 2020, 37(5): 151-155)
[4] PINHEIRO P, COLLOBERT R. Recurrent convolutional neural networks for scene labeling[C]//Proceedings of the 31st International Conference on International Conference on Machine Learning. Beijing: JMLR, 2014:82-90.
[5] 李丽双, 郭元凯. 基于CNN-BLSTM-CRF模型的生物医学命名实体识别[J]. 中文信息学报, 2018, 32(1): 116-122. (LI L S, GUO Y K. Biomedical named entity recognition with CNN-BLSTM-CRF[J]. Journal of Chinese information processing, 2018, 32(1): 116-122.)
[6] 韩普, 顾亮. 基于混合深度学习的中文医学实体抽取研究[J]. 图书情报工作, 2022, 66(14): 119-127. (HAN P, GU L. Research on extraction of Chinese medical entities based on hybrid deep learning[J]. Library and information service, 2022, 66(14): 119-127.)
[7] CHIU J P C, NICHOLS E. Named entity recognition with bidirectional LSTM-CNNs[J]. Transactions of the association for computational linguistics, 2016, 4: 357-370.
[8] 吴小雪, 张庆辉. 预训练语言模型在中文电子病历命名实体识别上的应用[J]. 电子质量, 2020(9): 61-65. (WU X X, ZHANG Q H. Application of pre-training language model in Chinese EMR named entity recognition[J]. Electronics quality, 2020(9): 61-65.)
[9] 陈仲永, 黄雍圣, 张旻, 等. 基于预训练模型的医药说明书实体抽取方法研究[J]. 计算机科学与探索, 2024, 18(7): 1911-1922. (CHEN Z Y, WANG Y S, ZHANG M, et al. A study on entity extraction method for pharmaceutical instructions based on pretrained models[J]. Journal of frontiers of computer science and technology, 2024, 18(7): 1911-1922.)
[10] 段宇锋, 贺国秀. 面向中文医学文本命名实体识别的神经网络模块分解分析[J]. 数据分析与知识发现, 2023, 7(2): 26-37. (DUAN Y F, HE G X. Analysis of neural network modules for named entity recognition of Chinese medical texts[J]. Data analysis and knowledge discovery, 2023, 7(2): 26-37.)
[11] 张芳丛, 秦秋莉, 姜勇, 等. 基于RoBERTa-WWM-BiLSTM-CRF的中文电子病历命名实体识别研究[J]. 数据分析与知识发现, 2022, 6(Z1): 251-262. (ZHANG F C, QIN Q L, JIANG Y, et al. Named entity recognition for Chinese EMR with RoBERTa-WWM-BiLSTM-CRF[J]. Data analysis and knowledge discovery, 2022, 6(Z1): 251-262.)
[12] 张云秋, 汪洋, 李博诚. 基于RoBERTa-wwm动态融合模型的中文电子病历命名实体识别[J]. 数据分析与知识发现, 2022, 6(Z1): 242-250. (ZHANG Y Q, WANG Y, LI B C. Identifying named entities of Chinese electronic medical records based on RoBERTa-wwm dynamic fusion model[J]. Data analysis and knowledge discovery, 2022, 6(Z1): 242-250.)
[13] 车万翔, 窦志成, 冯岩松, 等. 大模型时代的自然语言处理：挑战、机遇与发展[J]. 中国科学：信息科学, 2023, 53(9): 1645-1687. (CHE W X, DOU Z C, FENG Y S, et al. Towards a comprehensive understanding of the impact of large language models on natural language processing: challenges, opportunities and future directions[J]. Scientia sinica(informationis), 2023, 53(9): 1645-1687.)
[14] BROWN T, MANN B, RYDER N, et al. Language models are few-shot learners[J]. Advances in neural information processing systems, 2020, 33:1877-1901.
[15] GIRAY L. Prompt engineering with ChatGPT: a guide for academic writers[J]. Annals of biomedical engineering, 2023, 51(12): 2629-2633.
[16] HESTON T F, KHUN C. Prompt engineering in medical education[J]. International medical education, 2023, 2(3): 198-205.
[17] 赵晓伟, 祝智庭, 沈书生. 教育提示语工程：构建数智时代的认识论新话语[J]. 中国远程教育, 2023, 43(11): 22-31. (ZHAO X W, ZHU Z T, SHEN S S. Educational prompt engineering: constructing a new epistemological discourse in the era of digital intelligence[J]. Chinese journal of distance education, 2023, 43(11): 22-31.)
[18] WANG S H, SUN X F, LI X Y, et al. GPT-ner: named entity recognition via large language models[EB/OL]. [2024-11-25]. https://arxiv.org/abs/2304.10428.
[19] ZHOU H X, AUSTIN R, LU S C, et al. Complementary and integrative health information in the literature: its lexicon and named entity recognition[J]. Journal of the American Medical Informatics Association, 2024, 31(2): 426-434.
[20] HU Y, CHEN Q Y, DU J C, et al. Improving large language models for clinical named entity recognition via prompt engineering[J]. Journal of the American Medical Informatics Association, 2024, 31(9): 1812-1820.
[21] MONAJATIPOOR M, YANG J X, STREMMEL J, et al. LLMs in biomedicine: a study on clinical named entity recognition[EB/OL]. [2024-11-25]. https://arxiv.org/abs/2404.07376.
[22] 唐晓晟, 程琳雅, 张春红, 等. 大语言模型在学科知识图谱自动化构建上的应用[J]. 北京邮电大学学报(社会科学版), 2024, 26(1): 125-136. (TANG X S, CHENG L Y, ZHANG C H, et al. Application of large language models in automated construction of knowledge graphs for university subject domains[J]. Journal of Beijing University of Posts and Telecommunications(social sciences edition), 2024, 26(1): 125-136.)
[23] MA M D, TAYLOR A K, WANG W et al. DICE: data-efficient clinical event extraction with generative models[EB/OL]. [2024-11-25]. https://arxiv.org/abs/2208.07989.
[24] ASHOK D, LIPTON Z C. PromptNER: prompting for fewShot named entity recognition[EB/OL]. [2024-11-25]. https://arxiv.org/abs/2305.15444.
[25] 中华医学会内分泌学分会. 中国成人2型糖尿病胰岛素促泌剂应用的专家共识[J]. 中华内分泌代谢杂志, 2012, 28(4): 261-265. (Chinese Society of Endocrinology. Expert consensus on the use of insulin secretagogues in Chinese adults with type 2 diabetes[J]. Chinese journal of endocrinology and metabolism, 2012, 28(4): 261-265.)
[26]中华医学会糖尿病学分会.中国2型糖尿病防治指南(十五)——儿童和青少年糖尿病[J]. 中国社区医师, 2012, 28(9): 9. (Chinese Diabetes Society. Guidelines for the prevention and treatment of type 2 diabetes in China (XV): diabetes in children and adolescents[J]. Chinese community doctors, 2012, 28(9): 9.)
Options
文章导航

/

〈 〉