中国科技术语 ›› 2025, Vol. 27 ›› Issue (4): 136-141.doi: 10.12339/j.issn.1673-8578.2025.04.031

• 数据技术 • 上一篇    下一篇

基于大语言模型的企业术语提取实践

高志军1(), 季陈1(), 刘可2,*()   

  1. 1 北京大学软件与微电子学院,北京 102600
    2 小米科技有限责任公司,北京 100085
  • 收稿日期:2025-02-24 出版日期:2025-07-06 发布日期:2025-07-06
  • 通讯作者:
    *刘可,小米科技有限责任公司内容管理办公室负责人。通信方式:
  • 作者简介:

    高志军,博士,北京大学软件与微电子学院讲师、硕士生导师,研究方向为自然语言处理、用户体验、技术传播。通信方式:

    季陈,北京大学软件与微电子学院硕士研究生,研究方向为自然语言处理、数据挖掘、用户体验。通信方式:

  • 基金资助:
    北大-小米校企合作项目“基于大语言模型的术语提取”(MI113020230300002X)

The Practice of Corporate Terminology Extraction Based on Large Language Models

GAO Zhijun1(), JI Chen1(), LIU Ke2,*()   

  • Received:2025-02-24 Online:2025-07-06 Published:2025-07-06

摘要:

文章以小米公司术语为例,探讨了大语言模型在企业术语提取中的应用。通过实验分析了现有大模型在提取企业术语方面的能力,并结合提示词优化、微调模型等方法尝试增强其对企业术语的适应性。实验结果表明,设计合适的提示词能够提高企业术语提取的准确率,微调大模型也能够显著提升术语提取的准确性,为术语提取任务提供了实践经验。研究也为大模型在企业知识管理中的应用提供了可借鉴的思路与方法。

关键词: 大语言模型, 企业术语提取, 提示词工程

Abstract:

This paper takes corporate terminology from Xiaomi as an example to explore the application of large language models in tasks like corporate terminology extraction.Through experiments,the paper analyzes the ability of available large models in extracting corporate terminology,attempting to enhance their adaptability to corporate terms by methods like optimizing prompt words and fine-tuning models.The experimental results show that designing appropriate prompts can significantly improve the accuracy of corporate terminology extraction,so does fine-tuning,providing valuable practical experience for related tasks.The research also offers actionable insights and methods for the application of large models in knowledge management of enterprises.

Key words: large language models, corporate terminology extraction, prompt engineering