首页|期刊导航|指挥控制与仿真|面向演训文本数据的杀伤链信息提取方法

面向演训文本数据的杀伤链信息提取方法OA

A methodology for extracting kill chain information from military exercise text data

中文摘要英文摘要

作为反映演训情况的关键要素,军事演训活动产生的文本数据蕴含着评估作战效能所需的杀伤链关键信息.该类数据具有专业术语复杂、关键信息密集等特点.传统人工数据提取方法效率低且周期长,本文提出了一种基于微调垂直领域大模型的杀伤链信息提取方法:在微调阶段引入军事场景语料库,设计了针对杀伤链特征的提示词,并增加回答中高频关键词出现时的指标得分,弥补了通用大模型针对性不强的缺陷.实验结果表明,相较于通用大模型,微调后模型在杀伤链信息提取任务中的输出准确率提升 49%,输出质量提升 90%,为面向演训文本的杀伤链信息提取提供了有效可靠的技术途径.

As a key element reflecting the military exercise,the text data generated by military exercise contains the crucial information needed for evaluating combat effectiveness in the kill chain.This type of data has the characteristics of complex professional terms and dense key information.The traditional manual data extraction methods are inefficient and time-consu-ming,while the automated extraction methods using general large models are not targeted enough.This paper proposes a kill chain information extraction method based on fine-tuning a vertical domain large model:introducing a military scene corpus during the fine-tuning stage,designing prompt words specific to the kill chain features,and increasing the index score when high-frequency keywords appear in the answers,which compensates for the lack of targeting of general large models.Experi-mental results show that compared with general large models,the fine-tuned model has an output accuracy rate of 49%im-provement and a quality improvement of 90%in the kill chain information extraction task,providing an effective and reliable technical approach for the extraction of kill chain information from training and exercise texts.

高鑫;初阳

江苏自动化研究所,江苏 连云港 222061江苏自动化研究所,江苏 连云港 222061

军事科技

大语言模型演训文本杀伤链提取

large language modeltraining documentskill chain extraction

《指挥控制与仿真》 2026 (5)

101-108,8

10.3969/j.issn.1673-3819.2026.05.014

评论