QD-RAG:一种问题分解驱动的迭代检索增强生成方法

doi:10.6040/j.issn.1672-3961.0.2025.125

摘要/Abstract

摘要： 针对现有检索增强生成方法处理需多跳推理复杂问题时,因缺乏对问题结构层次化解析,检索结果与用户需求存在语义偏差问题,提出问题分解驱动的迭代检索增强生成(query decomposition driven iterative retrieval augmented generation,QD-RAG)方法。通过问题分解模块结构化处理用户查询,单跳问题用链式推理策略多维度解析,多跳问题逐层化分解;由大语言模型驱动的迭代检索模块进行多轮上下文检索,提升检索知识精度,生成高质量回答。试验结果表明,QD-RAG方法有效降低生成过程中幻觉现象,且在单跳、多跳及未知领域问答任务准确率上显著提升。

关键词: 检索增强生成, 问题分解, 大语言模型, 迭代检索, 多跳推理

Abstract: To address the semantic mismatch between retrieval results and user needs in existing retrieval-augmented generation methods when handling complex questions requiring multi-hop reasoning, a query decomposition driven iterative retrieval augmented generation method was proposed, termed QD-RAG. This mismatch arised from the lack of hierarchical analysis of query structures in conventional RAG methods. In the proposed framework, a query decomposition module first processed user queries in a structured manner: single-hop questions were analyzed from multiple dimensions using a chain-of-thought strategy, whereas multi-hop questions were decomposed progressively in a layer-by-layer manner. Subsequently, an iterative retrieval module driven by large language models performed multiple rounds of contextual retrieval, which improved the precision of retrieved knowledge and enabled the generation of high-quality answers. Experimental results showed that QD-RAG effectively reduced hallucinations during generation and significantly improved accuracy in single-hop, multi-hop, and out-of-domain question-answering tasks.

Key words: rag, question decomposition, large language models, iterative retrieval, multi-hop

中图分类号:

TP181

陈颖鑫,陈振涵,林晓宇,蔡志铃,陈林峰,王李进. QD-RAG:一种问题分解驱动的迭代检索增强生成方法[J]. 山东大学学报 (工学版), 2026, 56(4): 75-83.

Chen Yingxin, Chen Zhenhan, Lin Xiaoyu, Cai Zhiling, Chen Linfeng, Wang Lijin. QD-RAG: Question decomposition driven iterative retrieval augmented generation[J]. Journal of Shandong University(Engineering Science), 2026, 56(4): 75-83.

参考文献

[1] Brown T, Mann B, Ryder N, et al. Language models are few-shot learners[J]. Advances in Neural Information Processing Systems, 2020, 33: 1877-1901.
[2] Kasai J, Sakaguchi K, Takahashi Y, et al. RealTime QA: what's the answer right now?[PP/OL]. V2.(2024-02-28)[2025-07-22]. https://arxiv.org/abs/2207.13332
[3] Mallen A, Asai A, Zhong V, et al. When not to trust language models: investigating effectiveness of parametric and non-parametric memories[C] //Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Toronto, Canada: ACL, 2023: 9802-9822.
[4] Hu E J, Shen Y L, Wallis P, et al. LoRA: low-rank adaptation of large language models[PP/OL]. V2.(2021-10-16)[2025-07-22]. https://doi.org/10.48550/arXiv.2106.09685
[5] Gao Y F, Xiong Y, Gao X Y, et al. Retrieval-augmented generation for large language models: a survey[PP/OL]. V5.(2024-03-27)[2025-07-22].https://doi.org/10.48550/arXiv.2312.10997
[6] Zellers R, Bisk Y, Schwartz R, et al. SWAG: a large-scale adversarial dataset for grounded commonsense inference [C] //Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, Belgium: ACL, 2018: 93-104.
[7] Izacard G, Grave E. Leveraging passage retrieval with generative models for open domain question answering [C] //Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Online: ACL, 2021: 874-880.
[8] Borgeaud S, Mensch A, Hoffmann J, et al. Improving language models by retrieving from trillions of tokens[C] //Proceedings of the 39th International Conference on Machine Learning. New York, USA: PMLR, 2022: 2206-2240.
[9] Lewis P, Perez E, Piktus A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[C] //Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: ACM, 2020: 9459-9474.
[10] Guu K, Lee K, Tung Z, et al. Retrieval augmented language model pre-training[C] //Proceedings of the 37th International Conference on Machine Learning. New York, USA: PMLR, 2020: 3929-3938.
[11] Lazaridou A, Gribovskaya E, Stokowiec W, et al. Internet-augmented language models through few-shot prompting for open-domain question answering[PP/OL]. V2.(2022-05-23)[2025-07-22]. https://doi.org/10.48550/arXiv.2203.05115
[12] Yang Z L, Qi P, Zhang S Z, et al. HotpotQA: a dataset for diverse, explainable multi-hop question answering[C] //In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Brussels, Belgium: ACL, 2018: 2369-2380.
[13] Ram O, Levine Y, Dalmedigos I, et al. In-context retrieval-augmented language models[J]. Transactions of the Association for Computational Linguistics, 2023, 11: 1316-1331.
[14] Shi F, Chen X Y, Misra K, et al. Large language models can Be easily distracted by irrelevant context[C] //Proceedings of the 40th International Conference on Machine Learning. New York, USA: PMLR, 2023: 31210-31227.
[15] Wang Y X, Chiew V. On the cognitive process of human problem solving[J]. Cognitive Systems Research, 2010, 11(1): 81-92.
[16] Khattab O, Santhanam K, Li X L, et al. Demonstrate-search-predict: composing retrieval and language models for knowledge-intensive NLP[PP/OL]. V2.(2023-01-23)[2025-07-22]. https://doi.org/10.48550/arXiv.2212.14024
[17] Sun Z Q, Wang X Z, Tay Y, et al. Recitation-augmented language models[PP/OL]. V2.(2023-02-16)[2025-07-22]. https://doi.org/10.48550/arXiv.2210.01296
[18] Yu W H, Iter D, Wang S H, et al. Generate rather than retrieve: large language models are strong context generators[PP/OL]. V3.(2023-01-25)[2025-07-22]. https://doi.org/10.48550/arXiv.2209.10063
[19] Ma X B, Gong Y Y, He P C, et al. Query Rewriting in Retrieval-Augmented Large Language Models[C] // In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singapore: ACL, 2023: 5303-5315.
[20] Chan C M, Xu C P, Yuan R B, et al. RQ-RAG: learning to refine queries for retrieval augmented generation[PP/OL].(2024-03-31)[2025-07-22]. https://doi.org/10.48550/arXiv.2404.00610
[21] Zhou D, Schärli N, Hou L, et al. Least-to-most prompting enables complex reasoning in large language models[PP/OL].(2022-05-21)[2025-07-22]. https://doi.org/10.48550/arXiv.2205.1062
[22] Wei J, Wang X Z, Schuurmans D, et al. Chain-of-thought prompting elicits reasoning in large language models[J]. Advances in Neural Information Processing Systems, 2022, 35: 24824-24837.
[23] Shi W J, Min S, Yasunaga M, et al. REPLUG: retrieval-augmented black-box language models[C] //Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Mexico City, Mexico: ACL, 2024: 8371-8384.
[24] Izacard G, Lewis P, Lomeli M, et al. Atlas: few-shot learning with retrieval augmented language models[J]. Journal of Machine Learning Research, 2023, 24(251): 1-43.
[25] Shao Z H, Gong Y Y, Shen Y L, et al. Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy[C] //Findings of the Association for Computational Linguistics: EMNLP. Singapore: ACL, 2023: 9248-9274.
[26] Yao S Y, Zhao J, Yu D, et al. ReAct: synergizing reasoning and acting in language models[PP/OL]. V3.(2023-03-10)[2025-07-22]. https://arxiv.org/abs/2210.03629
[27] Feng Z Y, Feng X C, Zhao D Z, et al. Retrieval-generation synergy augmented large language models[C] //ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing. Seoul: IEEE, 2023: 11661-11665.
[28] Wei S J, Zhang W W, Li Q S, et al. Semantic parsing for question answering over knowledge graphs[PP/OL].(2023-12-01)[2025-07-22]. https://doi.org/10.48550/arXiv.2401.06772
[29] Liu L H, Hill B, Du B X, et al. Conversational question answering with language models generated reformulations over knowledge graph[C] // Findings of the Association for Computational Linguistics: Bangkok, Thailand: ACL, 2024: 839-850.
[30] Nguyen H H, Nguyen M T. Emotion-cause pair extraction as question answering[PP/OL]. V2.(2023-01-05)[2025-07-22]. https://doi.org/10.48550/arXiv.2301.01982
[31] 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: 453-466.
[32] Joshi M, Choi E, Weld D, et al. TriviaQA: a large scale distantly supervised challenge dataset for reading comprehension[C]. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics. Vancouver, Canada: ACL, 2017: 1601-1611.
[33] Ho X, Duong N A K, Sugawara S, et al. Constructing a multi-hop QA dataset for comprehensive evaluation of reasoning steps[J]. Proceedings of the 28th International Conference on Computational Linguistics. Barcelona, Spain: ACL, 2020: 6609-6625.
[34] Chen J W, Lin H Y, Han X P, et al. Benchmarking large language models in retrieval-augmented generation[C]. Proceedings of the 38th AAAI Conference on Artificial Intelligence,Vancouver, Canada: AAAI, 2024: 17754-17762.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed