计算机科学

计算机科学 ›› 2026, Vol. 53 ›› Issue (8): 285-297.doi: 10.11896/jsjkx.250600116

• 人工智能 • 上一篇    下一篇

基于检索任务规划与验证反思机制的AgenticRAG方法

张浩然, 郝文宁, 靳大尉, 程恺, 刘君阳   

  1. 陆军工程大学指挥控制工程学院 南京 210000
  • 收稿日期:2025-06-17 修回日期:2026-09-30 发布日期:2026-08-17
  • 通讯作者: 郝文宁(hwnbox@163.com)
  • 作者简介:(ycdfzhr919@163.com)
  • 基金资助:
    国家自然科学基金(61806221);国防基础科研计划(JCKY2020601b018)

Agentic Retrieval Augmented Generation Framework Based on Retrieval Task Planning and Reflection Mechanism

ZHANG Haoran, HAO Wenning, JIN Dawei, CHENG Kai, LIU Junyang   

  1. College of Command & Control Engineering, Army Engineering University of PLA, Nanjing 210000, China
  • Received:2025-06-17 Revised:2026-09-30 Online:2026-08-17
  • About author:ZHANG Haoran,born in 2001,postgraduate.His main research interests include natural language processing and data mining.
    HAO Wenning,born in 1971,Ph.D,professor,Ph.D supervisor.His main research interests include data mining and machine learning.
  • Supported by:
    National Natural Science Foundation of China(61806221) and National Defense Basic Research Program(JCKY2020601b018).

PDF (PC)

141

摘要： 大语言模型(Large Language Models,LLMs)在处理知识密集型任务时,幻觉与知识时效性问题制约了其在高可靠性领域的应用。传统检索增强生成(Retrieval Augmented Generation,RAG)系统通过调用外部知识帮助LLMs拓宽知识边界,但其在多跳推理任务中存在语义关联不足、检索策略固化等局限。为突破上述瓶颈,提出一种基于检索任务规划与验证反思机制的智能体检索增强生成框架(PR-RAG)。该框架通过任务规划模块对问题进行层次化分类与子问题分解,并结合向量检索与图检索的自适应混合检索策略获取上下文信息,最后通过逻辑验证与事实验证的双重验证机制形成闭环优化。实验结果表明,PR-RAG在HotpotQA和2WikiMQA等数据集上的精准匹配率性能指标较基线方法平均提升12.8%,有效提升了复杂知识密集型任务处理的准确性与鲁棒性。研究证实,分层检索任务规划与验证反思机制的协同反馈可显著提高LLMs的知识推理能力和系统可靠性。

关键词: 大语言模型, 检索增强生成, 智能体, 问答系统, 知识图谱

Abstract: When large language models(LLMs) handle knowledge-intensive tasks,issues of hallucinations and knowledge timeliness restrict their application in high-reliability fields.Traditional RAG(Retrieval-Augmented Generation) systems help LLMs expand their knowledge boundaries by invoking external knowledge;however,they suffer from limitations such as insufficient semantic relevance and rigid retrieval strategies in multi-hop reasoning tasks.To break through the aforementioned bottlenecks,this study proposes an intelligent iterative retrieval-augmented generation framework(PR-RAG) based on retrieval task planning and a verification-reflection mechanism.Specifically,this framework employs a task planning module to conduct hierarchical classification of questions and decomposition of sub-problems,acquires contextual information via an adaptive hybrid retrieval strategy that integrates vector retrieval and graph retrieval,and ultimately realizes closed-loop optimization through a dual verification mechanism(including logical verification and factual verification).Experimental results show that the exact matching of PR-RAG on datasets such as HotpotQA and 2WikiMQA are increased by an average of 12.8% compared with baseline methods,which effectively improves the accuracy and robustness in handling complex knowledge-intensive tasks.This study confirms that the synergistic feedback between hierarchical retrieval task planning and the verification-reflection mechanism can significantly enhance the knowledge reasoning capabilities of LLMs and the reliability of the system.

Key words: Large language models, Retrieval-augmented generation, Intelligent agent, Question-answering system, Knowledge graph

中图分类号: 

  • TP391
[1] GUO D Y,YANG D J,ZHANG H W,et al.DeepSeek-R1:Incentivizing reasoning capability in LLMs via reinforcement learning[R].Shenzhen:DeepSeek-AI,2025.
[2] ACHIAM J,ADLER S,AGARWAL S,et al.GPT-4 Technical Report[R].OpenAI,2023.
[3] ASAI A,MIN S,ZHONG Z.Self-RAG:Learning to Retrieve,Generate,and Critique through Self-Reflection[C]//Twelfth International Conference on Learning Representations.2023:41-46.
[4] ZHAO W X,ZHOU K,LI J,et al.A Survey on Evaluation of Large Language Models[J].Association for Computing Machinery,2024,39:2157-6904.
[5] LEWIS P,PEREZ E,PIKTUS A,et al.Retrieval-augmentedgeneration for knowledge-intensive nlp tasks[C]//Advances in Neural Information Processing Systems 33.2020:9459-9474.
[6] CHAN C M,XU C,YUAN R,et al.RQ-RAG:Learning to refine queries for retrieval augmented generation[J].arXiv:2501.00332-v1,2025.
[7] ZHU Y H,REN C Y,XIE S Y,et al.EMERGE:EnhancingMultimodal Electronic Health Records Predictive Modeling with Retrieval-Augmented Generation[C]//Proceedings of the 33rd ACM International Conference on Information and Knowledge Management.2024:3549-3559.
[8] SHI F,CHEN X Y,MISRA K,et al.Large language models can be easily distracted by irrelevant context[C]//In International Conference on Machine Learning(ICML 2024).PMLR,2024:31210-31227.
[9] GUO T C,CHEN X Y,WANG Y Q,et al.Large language mo-del based multi-agents:A survey of progress and challenges[C]//Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence(IJCAI ’24).2024:8048-8057.
[10] LU J Q,ZHONG W J,HUANG W Y,et al.SELF:Self-Evolution with Language Feedback[J].arXiv:2310.00533,2023.
[11] HUANG W L,ABBEEL P,PATHAK D,et al.Language Models as Zero-Shot Planners:Extracting Actionable Knowledge for Embodied Agents[C]//Proceedings of the 39th International Conference on Machine Learning.2022:9118-9147.
[12] HONG S R,ZHU GE M C,CHEN J Q,et al.MetaGPT:Merging Large Language Models Using Model Exclusive Task Arithmetic[C]//2024 Conference on Empirical Methods in Natural Language Processing.Association for Computational Linguistics,2024:1711-1724.
[13] SHINN N,CASSANO F,GOPINATH A,et al.Reflexion:Language Agents with Verbal Reinforcement Learning[C]//Conference on Neural Information Processing Systems.NeurIPS Foundation,377:8634-8652.
[14] ZHANG W L,LI X Y,DONG K C,et al.Process vs.Outcome Reward:Which is Better for Agentic RAG Reinforcement Learning[J].arXiv:2505.14069,2025.
[15] HUANG Y Z,HUANG J.A Survey on Retrieval-AugmentedText Generation for Large Language Models[J].arXiv:2404.10981,2024.
[16] LIANG L,SUN M S,GUI Z K,et al.KAG:Boosting LLMs in Professional Domains via Knowledge Augmented Generation[C]//EMNLP 2024.2024:1711-1724.
[17] EDGE D,TRINH H,CHENG N,et al.From Local to Global:A Graph RAG Approach to Query-Focused Summarization[J].arXiv:2404.16130,2024.
[18] SARMAH B,HALL B,RAO R,et al.HybridRAG:Integrating Knowledge Graphs and Vector Retrieval Augmented Generation for Efficient Information Extraction[C]//Proceedings of the 5th ACM International Conference on AI in Finance.ACM2024(ICAIF’24).2024:608-616.
[19] TANG X Q,GAO Q,LI J,et al.MBA-RAG:a Bandit Approach for Adaptive Retrieval-Augmented Generation through Question Complexity[C]//International Conference on Computational Linguistics.ACL2025,2025:3248-3254.
[20] CUCONASU F,TRAP-POLINI G,SICILIANO F,et al.Thepower of noise:Redefining retrieval for rag systems[C]//Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval.2024:719-729.
[21] TALEBIRAD Y,NADIRI A.Multi-agent collaboration:Har-nessing the power of intelligent LLM agents[J].arXiv:2306.03314,2023.
[22] SALVE A,ATTAR S,DESHMUKH M,et al.A Collaborative Multi-Agent Approach to Retrieval-Augmented Generation Across Diverse Data[J].arXiv:2412.05838,2024.
[23] CHANG C Y,JIANG Z M,RAKESH V,et al.MAIN-RAG:Multi-Agent Filtering Retrieval-Augmented Generation[J].arXiv:2501.00332v1,2025.
[24] DU S H,ZHAO J B,SHI J X,et al.A Survey on the Optimization of Large Language Model-based Agents[J].arXiv:2503.12434v1,2025.
[25] ZHOU J W,CHEN L.OpenRAG:Optimizing RAG End-to-End via In-Context Retrieval Learning[J].arXiv:2503.08398,2025.
[26] CHEN B,SHU C,SHAREGHI E,et al.FireAct:Toward Language Agent Fine-tuning[J].arXiv:2310.05915,2023.
[27] SCHULMAN J,WOLSKI F,DHARIWAL P,et al.ProximalPolicy Optimization Algorithms[R].OpenAI,2023.
[28] BARTO A G,SUTTON R S,ANDERSON C.Neuron like elements that can solve difficult learning control problems[J].IEEE Transactions on Systems,Man,& Cybernetics,1970,13(5):834-846.
[29] OUYANG L,WU J,JIANG X,et al.Training language models to follow instructions with human feedback[C]//36th Confe-rence on Neural Information Processing Systems.NeurIPS 2022,2022:15240-15253.
[30] KWIATKOWSKI T,PALOMAKI J,REDFIELD O,et al.Natural Questions:A Benchmark for Question Answering Research[J].Transactions of the Association for Computational Linguistics,2019,7(15):452-466.
[31] YANG Z L,QI P,ZHANG S Z,et al.HotpotQA:A Dataset for Diverse,Explainable Multi-hop Question Answering[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing.2018:2369-2380.
[32] YANG A,LI A F,YANG B S,et al.Qwen3 Technical Report[R].Qwen Team,2025.
[33] CHEN Y Q,YAN L Y,SUN W W,et al.Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning[J].arXiv:2501.00332v1,2025.
[34] ZHENG R,DOU S H,GAO S Y,et al.Secrets of RLHF in large language models part1:PPO[R].Fudan NLP Group,2023.
[35] SCHULMAN J,MORITZ P,LEVINE S,et al.High-dimen-sional continuous control using generalized advantage estimation[C]//4th International Conference on Learning Representations(ICLR 2016).2016:571-584.
[36] HO X,NGUYEN A K,et al.Constructing a Multi-Hop QA Dataset for Comprehensive Evaluation of Reasoning Steps[C]//28th International Conference on Computational Linguistics.COLING 2020,2020:580-589.
[37] MANAS G,CLARK J H,LEE K,et al.TriviaQA:A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension[C]//NAACL-HLT.NAACL,2018:1601-1611.
[38] TOUVRON H,MARTIN L,STONE K,et al.Llama 2:Open foundation and fine-tuned chat models[R].Meta AI,2023.
[39] GRATTAFIORI A,DUBEY A,JAUHRI A,et al.The Llama 3 Herd of Models[R].Meta AI,2024.
[40] MA X B,GONG Y Y,HE P C,et al.Query Rewriting for Retrieval-Augmented Large Language Models[C]//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing.2023:5303-5315.
[41] KE Z X,KONG W Z,LI C,et al.Bridging the Preference Gap between Retrievers and LLMs[C]//62nd Annual Meeting of the Association for Computational Linguistics(ACL 2024).2024:1523-1535.
[42] LIU Y M,PENG X Y,ZHANG X H,et al.RA-ISF:Learning to Answer and Understand from Retrieval Augmentation via Iterative Self-Feedback[C]//Association for Computational Linguistics 2024.2024:4730-4749.
[43] JEONG S Y,BAEK J H,CHO S M,et al.Adaptive-rag:Lear-ning to adapt retrieval-augmented large language models through question complexity[C]//Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies(NAACL 2024).2024:7036-7050.
[44] GAN A R,YU H,ZHANG K,et al.Retrieval Augmented Ge-neration Evaluation in the Era of Large Language Models:A Comprehensive Survey[J].arXiv:2504.14891,2025.
[45] IZACARD G,LEWIS P,LOMELI M,et al.Atlas:Few-shotlearning with retrieval augmented language models[J].Journal of Machine Learning Research,2023,24:1-43.
[46] KARPUKHIN V,OGUZ B,MIN S,et al.Dense passage retrie-val for open-domain question answering[C]//Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing2020.2020:6769-6781.
[47] VRANDEČIĆ D,KRÖT-ZSCH M.Wikidata:A Free Collabora-tive Knowledge Base[J].Communications of the ACM,2014,57(10):78-85.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市两江新区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发