数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (5): 184-197.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.05.015

doi: 10.11871/jfdc.issn.2096-742X.2025.05.015

• 技术与应用 • 上一篇    下一篇

基于提示学习的雷达领域知识抽取方法

黄振铭(),吴晓芳*(),薛孟武   

  1. 中国人民解放军32802部队,北京 100191
  • 收稿日期:2025-01-17 出版日期:2025-10-20 发布日期:2025-10-23
  • 通讯作者: 吴晓芳
  • 作者简介:黄振铭,中国人民解放军32802部队,硕士研究生,主要研究方向为雷达装备知识图谱。
    本文负责论文构思与主体内容撰写,搜集数据进行实验。
    HUANG Zhenming, is a master’s student at Unit 32802 of the People’s Liberation Army. His main research interest is knowledge graph of radar equipment.
    In this paper, he is responsible for the conception of the study, writing the main content of the manuscript, and collecting data for experiments.
    E-mail: 1213815602@qq.com|吴晓芳,中国人民解放军32802部队高工,主要研究方向为雷达信号处理、辐射源信号分析、电磁信息多源异构融合处理等。
    本文负责雷达知识专业理解与经验指导,负责论文整体框架、方向审定与把握。
    WU Xiaofang, is currently a senior engineer at Unit 32802 of the People's Liberation Army. Her main research interests include radar signal processing, emitter signal analysis,and heterogeneous multi-source fusion processing of electromagnetic information.
    In this paper, she is responsible for providing professional expertise in radar knowledge and guidance, as well as overall framework of the paper and the direction of the validation and grasp.
    E-mail: 542771912@qq.com

Method On Knowledge Extraction of Radar Domains Based on Prompt Learning

HUANG Zhenming(),WU Xiaofang*(),XUE Mengwu   

  1. 32802 of the People’s Liberation Army Unit, Beijing 100191, China
  • Received:2025-01-17 Online:2025-10-20 Published:2025-10-23
  • Contact: WU Xiaofang

摘要:

【目的】 雷达领域的二重信息是对雷达装备性能更清晰完整的描述,针对此类问题的研究较少,本文提出一种基于提示学习和LoRA微调大模型的雷达装备领域知识抽取方法(Radar Knowledge Extraction based on Prompt Learning and LoRA fine-tuning LLM, RKEPLL)。【方法】 设计了雷达装备领域的提示学习模板及关键词组合,将二重信息抽取建模成信息组合生成,对LLaMA3进行LoRA微调,同时利用公众号及雷达手册中的数据构建了领域数据集对该方法进行了验证。【结果】 在自建数据集(2,800条数据,近120万字)上取得了不错的效果,其BLEU-4、ROUGE-1、ROUGE-2、ROUGE-L分别为97.1406%、99.2544%、99.0758%、97.9330%,且具备较好的实际应用能力,为解决雷达装备领域的二重信息抽取提供了一种可行的方法。

关键词: 提示学习, 信息组合生成, 二重信息抽取, 雷达装备

Abstract:

[Objective] Dual information in the field of radar provides a clearer and more complete description of the performance of radar equipment, but fewer studies have been conducted on such problems. In this paper, a radar knowledge extraction method based on Prompt Learning and LoRA fine-tuning LLM (RKEPLL) is proposed for knowledge extraction in the field of radar equipment. [Method] A prompt learning template and keyword combination in the field of radar equipment are designed to model dual information extraction into information combination generation. In addition, LoRA fine-tuning is performed on LLaMA3, and the method is also validated by constructing a domain dataset using data from the WeChat Official Accounts and radar manuals. [Conclusion] Efficient results have been achieved on the self-constructed dataset, whose BLEU-4, ROUGE-1, ROUGE-2, and ROUGE-L are 97.1406%, 99.2544%, 99.0758% and 97.9330%, respectively, and have certain practical application ability, which provides an efficient method for solving the dual information extraction in the field of radar equipment.

Key words: prompt learning, information combination generation, dual information extraction, radar equipment