罗一雄,滕尚志,吕学强,游新冬

• 1:北京信息科技大学网络文化与数字传播北京市重点实验室
• 2:北京印刷学院信息工程学院

摘要(Abstract)：

针对通用大语言模型在军事、医疗、金融等专业问答场景中幻觉严重、专业性不足的问题，以及检索增强生成（retrieval-augmented generation,RAG）技术缓解模型幻觉过程中易受检索文档虚假信息误导的新问题，提出一种可信度感知多策略图RAG(credibility-aware multistrategy graph RAG,CAMG-RAG)框架，构建“可信知识生成—高可信信息检索—注意力权重优化”三级协同架构。首先，通过改进的知识图谱链接器生成含实体、关系、三元组及可信度评分的图构件；其次，基于可信度评估机制，实现多策略图检索；最后，定位大模型中对知识图谱元素敏感的“有影响力注意力头”，依据可信度评分修正其注意力权重，引导模型优先聚焦高价值知识。实验结果表明，在军事领域QAonMilitaryKG数据集上，CAMG-RAG能够显著提高大模型对抗幻觉的能力，基于Qwen3-4B模型，其忠实度、答案正确性分别达到0.746、0.863，比微调大模型提高14.6%、5.9%。

关键词(KeyWords)： 大语言模型;检索增强生成(retrieval-augmented generation,RAG);问答系统;知识图谱

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金项目(62171043);; 国防科技重点实验室基金项目(6412006200404)

作者(Author): 罗一雄,滕尚志,吕学强,游新冬

DOI: 10.16508/j.cnki.11-5866/n.2026.02.005

参考文献(References)：

• [1]ACHIAM J,ADLER S,AGARWAL S,et al. GPT-4 technical report[EB/OL].(2024-03-04)[2025-09-03]. https：//arxiv. org/abs/2303. 08774.
• [2]GUO D Y,YANG D J,ZHANG H W,et al. DeepSeek-R1:incentivizing reasoning capability in LLMs via reinforcement learning[EB/OL].(2025-01-22)[2025-09-10]. https：//arxiv.org/abs/2501. 12948.
• [3]LI Y W,WANG C Y,JIA J Y. LLaMA-VID:an image is worth 2tokens in large language models[C]//Computer Vision-ECCV2024. Cham:Springer,2025:323-340.
• [4]何静，沈阳，谢润锋.大语言模型幻觉现象的分类识别与优化研究[J].计算机科学与探索，2025,19(5):1295-1301.HE J,SHEN Y,XIE R F. Research on categorical recognition and optimization of hallucination phenomenon in large language models[J]. Journal of Frontiers of Computer Science&Technology,2025,19(5):1295-1301.(in Chinese)
• [5]刘泽垣，王鹏江，宋晓斌，等.大语言模型的幻觉问题研究综述[J].软件学报，2025,36(3):1152-1185.LIU Z Y,WANG P J,SONG X B,et al. Survey on hallucinations in large language models[J]. Journal of Software,2025,36(3):1152-1185.(in Chinese)
• [6]YU Y,PING W,LIU Z H,et al. RankRAG:unifying context ranking with retrieval-augmented generation in LLMs[J].Advances in Neural Information Processing Systems,2025,37:121156-121184.
• [7]WAN Y W,CHEN Z Y,LIU Y,et al. Empowering LLMs by hybrid retrieval-augmented generation for domain-centric Q&A in smart manufacturing[J]. Advanced Engineering Informatics,2025,65(Part B):103212.
• [8]ASAI A,WU Z Q,WANG Y Z,et al. Self-RAG:learning to retrieve,generate,and critique through self-reflection[C]//The12th International Conference on Learning Representations.Vienna:ICLR,2024:1-30.
• [9]HOFFMANN J,BORGEAUD S,MENSCH A,et al. Training compute-optimal large language models[C]//36th Conference on Neural Information Processing Systems(NeurIPS). La Jolla,California, USA:NIPS, 2022:1-36.
• [10]秦董洪，李政韬，白凤波，等.大语言模型参数高效微调技术综述[J].计算机工程与应用，2025,61(16):38-63.QIN D H,LI Z T,BAI F B,et al. Review of parameter-efficient fine-tuning technology for large language models[J]. Computer Engineering and Applications,2025,61(16):38-63.(in Chinese)
• [11]CHEN S Y,HOU Y T,CUI Y M,et al. Recall and learn:finetuning deep pretrained language models with less forgetting[C]//Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing(EMNLP). Stroudsburg, PA,USA:Association for Computational Linguistics,2020:7870-7881.
• [12]张川，卫晖，徐照兴，等.基于LoRA-SD模型实现民族服饰特征保留与创新的AIGC生成策略[J/OL].棉纺织技术，2025:1-10.(2025-10-10)[2025-10-22]. https：//doi. org/10. 26967/j. issn1000-7415. 202508003.ZHANG C,WEI H,XU Z X,et al. An AIGC generation strategy for achieving the preservation and innovation of ethnic costume features with the LoRA-SD model[J/OL]. Cotton Textile Technology,2025:1-10.(2025-10-10)[2025-10-22]. https：//doi. org/10. 26967/j. issn1000-7415. 202508003.(in Chinese)
• [13]张钦彤，王昱超，王鹤羲，等.大语言模型微调技术的研究综述[J].计算机工程与应用，2024,60(17):17-33.ZHANG Q T,WANG Y C,WANG H X,et al. Comprehensive review of large language model fine-tuning[J]. Computer Engineering and Applications,2024,60(17):17-33.(in Chinese)
• [14]PATIL R,HESTON T F,BHUSE V. Prompt engineering in healthcare[J]. Electronics,2024,13(15):2961.
• [15]ZHONG R,XU Y F,ZHANG C,et al. Leveraging large language model to generate a novel metaheuristic algorithm with CRISPE framework[J]. Cluster Computing,2024,27(10):13835-13869.
• [16]刘雪颖，云静，李博，等.基于大型语言模型的检索增强生成综述[J].计算机工程与应用，2025,61(13):1-25.LIU X Y,YUN J,LI B,et al. Survey of retrieval-augmented generation based on large language models[J]. Computer Engineering and Applications,2025,61(13):1-25.(in Chinese)
• [17]沈思，冯暑阳，吴娜，等.融合大语言模型的政策文本检索增强生成研究[J].数据分析与知识发现，2025,9(9):37-48.SHEN S,FENG S Y,WU N,et al. Retrieval-augmented generation of policy texts based on large language models[J].Data Analysis and Knowledge Discovery,2025,9(9):37-48.(in Chinese)
• [18]沈健，朱闽峰，马昱欣，等. KLMVis：基于知识图谱的检索增强语言模型可视分析系统[J/OL].计算机辅助设计与图形学学报，2024:1-12.(2024-10-29)[2025-05-14]. https：//link.cnki. net/urlid/11. 2925. tp. 20241029. 1415. 037.SHEN J,ZHU M F,MA Y X,et al. KLMVis:knowledge graphbased retrieval augmented language model visual analysis system[J/OL]. Journal of Computer-Aided Design&Computer Graphics,2024:1-12.(2024-10-29)[2025-05-14]. https：//link. cnki. net/urlid/11. 2925. tp. 20241029. 1415. 037(in Chinese)
• [19]DONG Y X,WANG S,ZHENG H Y,et al. Advanced RAG models with graph structures:optimizing complex knowledge reasoning and text generation[C]//2024 5th International Symposium on Computer Engineering and Intelligent Communications(ISCEIC). New York, USA:IEEE,2024:626-630.
• [20]刘焕勇.军事领域知识图谱问答系统（QAonMilitaryKG）[EB/OL].(2019-05-01)[2024-09-20]. https：//github. com/RomanGao/QAonMilitaryKG.LIU H Y. Military knowledge graph question answering system(QAonMilitaryKG)[EB/OL].(2019-05-01)[2024-09-20].https：//github. com/RomanGao/QAonMilitaryKG.(in Chinese)
• [21]刘焕勇.医疗知识图谱问答系统（QASystemOnMedicalKG）[EB/OL].(2018-10-05)[2024-10-15]. https：//github. com/liuhuanyong/QASystemOnMedicalKG.LIU H Y. Medical knowledge graph question Answering System(QASystemOnMedicalKG)[EB/OL].(2018-10-05)[2024-10-15]. https：//github. com/liuhuanyong/QASystemOnMedicalKG.(in Chinese)
• [22]ES S,JAMES T,ESPINOSA-ANKE L,et al. RAGAS:automated evaluation of retrieval augmented generation[EB/OL].(2025-04-28)[2025-09-10]. https：//arxiv. org/abs/2309. 15217.