面向低空经济政策多跳问答的知识抽取策略

洪文兴 ,  李宁 ,  李俊 ,  黄燕燕

天津大学学报(自然科学与工程技术版) ›› 2026, Vol. 59 ›› Issue (11) : 1197 -1207.

PDF (1691KB)
天津大学学报(自然科学与工程技术版) ›› 2026, Vol. 59 ›› Issue (11) : 1197 -1207. DOI: 10.11784/tdxbz202507029

面向低空经济政策多跳问答的知识抽取策略

作者信息 +

Knowledge Extraction Strategy for Multi-Hop Question Answering over Low-Altitude Economy Policies

Author information +
文章历史 +
PDF (1731K)

摘要

低空经济政策文本分析是典型的多跳问答任务,常面临推理链条复杂与长文本噪声干扰问题.为此,提出一种子问题引导的细粒度知识抽取策略(SG-FKE),通过查询重写将复杂问题分解为语义明确的子问题,引导模型聚焦核心意图,并从长文本中抽取用于多跳推理的结构化三元组,去除冗余内容,构建清晰的推理路径,从而提升问答系统的抗干扰能力与推理准确性.在公开多跳问答数据集 HotpotQA、2Wiki、MuSiQue 及自构建的低空经济政策问答数据集(LEMQA)上开展实验,实验结果显示,在 ChatGLM-9 B 模型上,SG-FKE 在 HotpotQA 的 F1 和 EM 分别达到 62.6%和 51.0%,在 2Wiki 上分别为 54.6%和 46.9%,在 MuSiQue 上分别为 48.8%和 29.9%,在 LEMQA 上 ROUGE 和 BERTScore-F1 分别为 44.6%和 73.3%;在 Qwen-32 B 模型上各项指标进一步提升,均显著优于基线方法.同时,实验还表明,SG-FKE 能够将原始上下文文本显著压缩至约 5%~15%的长度,有效减少冗余信息.此外,案例分析显示,SG-FKE 不仅在逻辑构建、语义聚焦与知识整合方面具备优势,还能够在含有时效性条款的政策问答中生成与时间相关的子问题,抽取关键时间三元组,从而避免答案失真,提升了结果的可靠性.消融实验进一步验证了子问题引导与三元组抽取机制在提升推理质量方面的关键作用.整体结果表明,SG-FKE 能够增强大语言模型对低空经济政策长文本的多跳推理能力与时效性条款处理能力,为智能化政策问答提供了一种轻量化、可推广的技术方案,具有重要的理论意义与实践价值.

Abstract

Analyzing low-altitude economy policy texts presents a typical multi-hop question answering(QA) challenge,often hindered by complex reasoning chains and noise from lengthy texts. To tackle these problems,a sub-question-guided fine-grained knowledge extraction strategy(SG-FKE)was developed. This strategy employed query rewriting to break down complex questions into clear,semantically focused subquestions,directing the model toward its core intent. Structured triples for multi-hop reasoning were extracted from extensive texts,redundant content was eliminated,and clear reasoning paths were constructed,thereby enhancing the robustness and accuracy of the QA system. Experiments were conducted on three public multi-hop QA datasets—HotpotQA ,2Wiki,and MuSiQue—and a self-constructed dataset , the low-altitude economy multi-hop question answering dataset(LEMQA). Results from the ChatGLM-9 B model show that SG-FKE achieves an F1 score of 62.6% and an exact match rate of 51.0% on HotpotQA;54.6% and 46.9% on 2Wiki;48.8% and 29.9% on MuSiQue;and ROUGE and BERTScore-F1 scores of 44.6% and 73.3% on LEMQA,respectively. All metrics further improve with the Qwen-32 B model,significantly outperforming baseline methods. In addition,our experiments show that SG-FKE can condense original context texts to approximately 5%—15% of their length,effectively minimizing redundancies. Case analyses demonstrate that SG-FKE excels in logical construction,semantic focus,and knowledge integration. It also generates time-related subquestions and extracts essential temporal triples in policy QA involving time-sensitive clauses,thus preventing answer distortion and increasing result reliability. Ablation studies confirm the importance of the sub-question guidance and triple extraction mechanisms in enhancing reasoning quality. Overall,our findings show that SG-FKE significantly improves the multi-hop reasoning and time-sensitive clause processing capabilities of large language models when working with lengthy low-altitude economy policy texts,thus offering a lightweight and generalizable technical solution for intelligent policy QA with significant theoretical and practical significance.

关键词

多跳问答 / 检索增强生成 / 知识抽取 / 低空经济政策

Key words

multi-hop question answering / retrieval-augmented generation / knowledge extraction / low-altitude economy policy

引用本文

引用格式 ▾
洪文兴,李宁,李俊,黄燕燕. 面向低空经济政策多跳问答的知识抽取策略[J]. 天津大学学报(自然科学与工程技术版), 2026, 59(11): 1197-1207 DOI:10.11784/tdxbz202507029

登录浏览全文

4963

注册一个新账户 忘记密码

参考文献

[1]

刘先江, 宋 丹, 徐 政. 以低空经济打造新质生产力发展新引擎[J]. 北京航空航天大学学报(社会科学版), 2024, 37(5): 134-144.

[2]

Liu Xianjiang, Song Dan, Xu Zheng. Creating a new engine for the development of new quality productive forces through low—altitude economy[J]. Journal of Beijing University of Aeronautics and Astronautics (Social Sciences Edition), 2024, 37(5): 134-144(in Chinese).

[3]

孔得建, 袁 泽. 低空经济政策法律体系的现状、经验与展望[J]. 北京航空航天大学学报(社会科学版), 2024, 37(5): 85-95.

[4]

Kong Dejian, Yuan Ze. Current situation,experience and prospect of the policies and legal system of low—altitude economy[J]. Journal of Beijing University of Aeronautics and Astronautics(Social Sciences Edition), 2024, 37(5): 85-95(in Chinese).

[5]

Kramer M A, Leavens A, Scarlat A. Harnessing AI for efficient analysis of complex policy documents:A case study of Executive Order 14110[EB/OL]. https://arxiv.org/abs/2406.06657, 2024—06—10.

[6]

Ren R, Wang Y, Qu Y, et al. Investigating the factual knowledge boundary of large language models with retrieval augmentation[C]// Proceedings of the 31st International Conference on Computational Linguistics. Abu Dhabi,UAE, 2025: 3697-3715.

[7]

Lewis P, Perez E, Piktus A, et al. Retrieval—augmented generation for knowledge—intensive NLP tasks [J]. Advances in Neural Information Processing Systems, 2020, 33: 9459-9474.

[8]

冯拓宇, 李伟平, 郭庆浪, 等. 大语言模型增强的知识图谱问答研究进展综述[J]. 计算机科学与探索, 2024, 18(11): 2887-2900.

[9]

Feng Tuoyu, Li Weiping, Guo Qinglang, et al. Overview of knowledge graph question answering enhanced by large language models[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(11): 2887-2900(in Chinese).

[10]

Kumar P.. Large language models(LLMs):Survey,technical frameworks,and future challenges[J]. Artificial Intelligence Review, 2024, 57(10): 260.

[11]

Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach,USA, 2017: 6000-6010.

[12]

Ji Z W, Lee N, Frieske R, et al. Survey of hallucination in natural language generation[J]. ACM Computing Surveys, 2023, 55(12): 1-38.

[13]

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.

[14]

Gao L, Ma X G, Lin J, et al. Precise zero—shot dense retrieval without relevance labels[C]// Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics(Volume 1:Long Papers). Toronto,Canada, 2023: 1762-1777.

[15]

Chan C M, Xu C P, Yuan R B, et al. RQ—RAG:Learning to refine queries for retrieval augmented generation[EB/OL]. https://arxiv.org/abs/2404.00610, 2024—03—31.

[16]

Sawarkar K, Mangal A, Solanki S R. Blended RAG:Improving RAG (retriever—augmented generation) accuracy with semantic search and hybrid query—based retrievers[EB/OL]. https://arxiv.org/abs/2404.07220, 2024—08—04.

[17]

Asai A, Wu Z Q, Wang Y Z, et al. Self—RAG:Learning to retrieve,generate,and critique through self—reflection[EB/OL]. https://arxiv.org/abs/2310.11511, 2023—10—17.

[18]

Zhang T J, Patil S G, Jain N, et al. Raft:Adapting language model to domain specific RAG[EB/OL]. https://arxiv.org/abs/2403.10131, 2024—03—15.

[19]

Fan W Q, Ding Y J, Ning L B, et al. A survey on RAG meeting LLMs: Towards retrieval—augmented large language models[C]// Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York,USA, 2024: 6491-6501.

[20]

Liu N F, Lin K, Hewitt J, et al. Lost in the middle:How language models use long contexts[J]. Transactions of the Association for Computational Linguistics, 2024, 12: 157-173.

[21]

Edge D,Trinh H,Cheng N, et al. From local to global:A graph RAG approach to query—focused summarization[EB/OL]. https://arxiv.org/abs/2404.16130 , 2024—04—24.

[22]

Wu J D, Zhu J Y, Qi Y L, et al. Medical graph RAG:Evidence—based medical large language model via graph retrieval—augmented generation[C]// Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics(Volume 1:Long Papers). Vienna,Austria, 2025: 28443-28467.

[23]

Yi P, Liang L, Zhang D, et al. KGFabric:A scalable knowledge graph warehouse for enterprise data interconnection[J]. Proceedings of the VLDB Endowment, 2024, 17(12): 3841-3854.

[24]

Liang L, Bo Z P, Gui Z K, et al. KAG:Boosting LLMs in professional domains via knowledge augmented generation[C]// Companion Proceedings of the ACM on Web Conference 2025. Sydney,Australia, 2025: 334-343.

[25]

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. Brussels,Belgium, 2018: 2369-2380.

[26]

Ho X, Nguyen A K D, Sugawara S, et al. Constructing a multi—hop QA dataset for comprehensive evaluation of reasoning steps[C]// Proceedings of the 28th International Conference on Computational Linguistics. Barcelona,Spain, 2020: 6609-6625.

[27]

Trivedi H, Balasubramanian N, Khot T, et al. MuSiQue:Multihop questions via single—hop question composition[J]. Transactions of the Association for Computational Linguistics, 2022, 10: 539-554.

基金资助

福建省创新战略研究计划资助项目(2024R0157)

AI Summary AI Mindmap
PDF (1691KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/

〈 〉