情报杂志2025，Vol.44Issue(7)：165-173,9.DOI:10.3969/j.issn.1002-1965.2025.07.020
国家安全情报战略知识图谱构建与检索增强问答框架研究
Research on Knowledge Graph Construction and Retrieval-Enhanced Question Answering Framework for National Security Intelligence Strategy
摘要
Abstract
[Research purpose]Against the backdrop of intensifying global strategic competition and profound geopolitical restructuring,constructing domain-specific knowledge systems from massive unstructured documents and conducting systematic analysis of national strat-egies has emerged as a critical research agenda in national security and open-source intelligence studies.[Research method]Taking U.S.national security intelligence strategy as an experimental corpus,this study proposes a knowledge graph construction method and retrieval-enhanced question answering framework.First,a fine-grained entity and relationship classification system is established,incorporating chain-of-thought and self-reflection prompting strategies to train a QLoRA-based professional knowledge extraction model adapted to na-tional security and strategic domains.Second,a structured and sustainably updatable knowledge graph is constructed based on a six-tuple knowledge representation scheme.Finally,a prompt-enhanced question answering framework integrating hybrid retrieval graphs and dy-namic expert role injection is developed to achieve precise matching and deep analysis of professional knowledge.[Research result/con-clusion]Experimental results demonstrate that the proposed method achieves Fl scores of 0.7181 and 0.7354 in entity recognition and re-lationship extraction tasks respectively,significantly outperforming baseline models.Experimental results demonstrate that our QA system achieved better performance than Claude-3.5-sonnet and GPT-o1 in both human evaluation and quantitative metrics.The research out-comes provide a reliable knowledge foundation and analytical tools for national security intelligence strategy research.关键词
国家安全情报/情报战略/大语言模型/知识图谱/检索增强生成/问答系统Key words
national security intelligence/intelligence strategy/large language models/knowledge graphs/retrieval-enhanced genera-tion/question-answering system引用本文复制引用
刘耀文,夏一雪,张鹏,沈宇航,韦昱妃,覃瑶..国家安全情报战略知识图谱构建与检索增强问答框架研究[J].情报杂志,2025,44(7):165-173,9.基金项目
公安部技术研究计划"社会安全事件网络舆情感知-推演-引导一体化风险防范技术研究"(编号:2023JSYC20) （编号:2023JSYC20）
教育部人文社会科学研究项目"智媒体时代网络舆情风险异常识别与治理研究"(编号:23YJCZH131)研究成果. （编号:23YJCZH131）
