图书情报工作

图书情报工作 >

2026 , Vol. 70 >Issue 18: 118 - 132

DOI: https://doi.org/10.13266/j.issn.0252-3116.2026.18.009

研究论文

PSR-CoT：一种面向学科交叉语义嵌套三元组抽取的提示框架

  • 肖玥 ,
  • 李艺亭
展开
  • 1 中国社会科学评价研究院 北京 100732;
    2 中国人民大学信息资源管理学院 北京 100872
肖玥，助理研究员，博士；李艺亭，博士研究生，通信作者，E-mail： ting2020@ruc.edu.cn。
32305.14.CN11-1541.2026.18.009

收稿日期: 2025-07-15

  修回日期: 2025-09-23

  网络出版日期: 2026-09-08

收起

PSR-CoT: A Prompting Framework for Interdisciplinary Nested Triplet Extraction

  • Xiao Yue ,
  • Li Yiting
Expand
  • 1 Chinese Academy of Social Sciences Evaluation Studies, Beijing 100732;
    2 School of Information Resource Management, Renmin University of China, Beijing 100872
Xiao Yue, Assistant Research Fellow, PhD; Li Yiting, Doctoral Candidate, Corresponding Author, E-mail: ting2020@ruc.edu.cn.

Received date: 2025-07-15

  Revised date: 2025-09-23

  Online published: 2026-09-08

Fold

摘要

[目的/意义] 为弥补现有学科交叉研究在语义层面结构化知识表征的不足，探索面向学科交叉核心内容的三元组描述模型及其抽取方法。[方法/过程] 首先，提出学科交叉“研究问题—解决方案—解决结果” PSR三元组描述框架，以实现对学科交叉研究中核心内容的结构化语义描述。其次，针对三元组抽取任务，设计基于大语言模型和思维链的PSR-CoT（problem-solution-result chain of thought）提示，通过在提示中引入文本描述特征、逻辑推理示例和自我反思机制等，有效提升大语言模型的三元组抽取精准性。[结果/结论] 基于自建数据集，引入本文提出的PSR-CoT提示策略后，Qwen3、GPT-4o和DeepSeek-R1在三元组抽取中F1值平均达到了86.11%、78.48%和70.01%，相比零提示策略分别提升了25.75%、18.61%和20.87%。Qwen3在抽取任务中表现最佳且与PSR-CoT提示配合最好，逻辑推理示例对模型性能提升贡献最大。基于PSR三元组构建的知识图谱，能够直观呈现在学科交叉问题解决过程中多学科知识的融合方式与路径，为相关研究提供技术路线参考。

关键词： 学科交叉三元组; 知识抽取; 提示策略; 思维链; 大语言模型; 科学问题视角

本文引用格式

肖玥 , 李艺亭 . PSR-CoT：一种面向学科交叉语义嵌套三元组抽取的提示框架[J]. 图书情报工作, 2026 , 70(18) : 118 -132 . DOI: 10.13266/j.issn.0252-3116.2026.18.009

Abstract

[Purpose/Significance] To address the lack of structured semantic representation in interdisciplinary research, this study explores a triplet-based descriptive model and its extraction method for capturing core interdisciplinary content. [Method/Process] This study proposes a semantic nested triplet of the form “Problem-Solution-Result” to enable a structured semantic representation of core interdisciplinary research content. To address the triplet extraction, a PSR-CoT (Problem-Solution-Result Chain-of-Thought) prompting strategy is designed based on large language models (LLMs) and the chain of thought (CoT). By incorporating characteristic descriptions, logical examples, and self-verification mechanisms into PSR-CoT, the triplet extraction accuracy is significantly improved. [Result/Conclusion] On a self-constructed dataset and the introduction of the PSR-CoT proposed in this paper, the average F1 scores of Qwen3, GPT-4o, and DeepSeek-R1 in triplet extraction reach 86.11%, 78.48%, and 70.01%, representing improvements of 25.75%, 18.61%, and 20.87% over zero-prompt baselines, respectively. Among these models, Qwen3 demonstrates the most effective synergy with PSR-CoT, achieving the highest performance. Logical examples contribute most significantly to the performance gain. The knowledge graphs constructed from these triplets visually map interdisciplinary problem-solving pathways and knowledge integration processes, offering a technical reference for related research.

Key words： interdisciplinary triplet; knowledge extraction; prompt strategy; chain of thought; large language model; research problem perspective

参考文献

[1] KLEIN J T, NEWELL W H. Advancing interdisciplinary studies[M] //GAFF J G, RATCLIFF J L. Handbook of the undergraduate curriculum: a comprehensive guide to the purposes, structures, practices, and change. San Francisco: Jossey-bass publishers, 1997: 393-394.
[2] WEI X, CUI X, CHENG N, et al. Chatie: zero-shot information extraction via chatting with ChatGPT[J/OL]. arXiv, 2023[2025-11-13]. https://arxiv.org/abs/2302.10205.
[3] QIAO B, ZOU Z, HUANG Y, et al. A joint model for entity and relation extraction based on BERT[J]. Neural computing and applications, 2022, 34:3471-3481.
[4] WAN Z, CHENG F, MAO Z, et al. GPT-re: in-context learning for relation extraction using large language models[J/OL]. arXiv, 2023[2025-05-29]. https://arxiv.org/abs/2305.02105.
[5] 徐春,苏明钰,马欢,等.基于大语言模型和提示学习的旅游文本实体关系联合抽取方法[J].数据分析与知识发现, 2025, 9(7): 130-140. (XU C, SU M Y, MA H, et al. Entity-relation joint extraction method for tourism text based on large language model and prompt learning[J]. Data analysis and knowledge discovery, 2025, 9(7): 130-140.)
[6] 刘启刚,王尹凡,牟立峰,等.大语言模型赋能“问题与方法”学术图谱构建与应用研究[J/OL].数据分析与知识发现, 1-16[2025-11-13]. https://hfgga60aabc7d15084b00h5v6p9kv66unq6xf5fhaz.libproxy.ruc.edu.cn/kcms/detail/10.1478. G2.20250207.1425.010.html. (LIU Q G, WANG Y F, MU L F, et al. Research on llm empowering construction and application of problem and method-oriented academic graph[J/OL]. Data analysis and knowledge discovery, 1-16[2025-05-29]. https://hfgga60aabc7d15084b00h5v6p9kv66unq6xf5fhaz.libproxy.ruc.edu.cn/kcms/detail/10.1478.G2.20250207.1425.010.html.)
[7] 程齐凯,李鹏程,张国标,等.学术文本词汇功能识别——基于标题生成策略和注意力机制的问题方法抽取[J].情报学报, 2021, 40(1): 43-52. (CHENG Q K, LI P C, ZHANG G B, et al. Recognition of lexical functions in academic texts: problem method extraction based on title generation strategy and attention mechanism[J]. Journal of the China Society for Scientific and Technical Information, 2021, 40(1): 43-52.)
[8] 张颖怡,章成志.基于公式化表达脱敏与边界识别加强的学术论文研究问题与方法识别研究[J].情报学报, 2024, 43(6): 712-732. (ZHANG Y Y, ZHANG C Z. Identification of problem and method in scientific papers based on formulaic expression desensitization and enhanced boundary recognition[J]. Journal of the China Society for Scientific and Technical Information, 2024, 43(6): 712-732.)
[9] 李雪思,张智雄,刘熠,等.科技文献研究问题句识别方法研究[J].图书情报工作, 2023, 67(9): 132-140. (LI X S, ZHANG Z X, LIU Y, et al. A study on the method of identifying research question sentences in scientific articles[J]. Library and information service, 2024, 43(6): 712-732.)
[10] 黄晓捷,熊回香,肖兵,等.基于“问题——方法”知识元挖掘的学科知识流动研究[J].图书情报工作, 2024, 68(8): 80-96. (HUANG X J, XIONG H X, XIAO B, et al. Research on subject knowledge flow based on “problem-method” knowledge units mining[J]. Library and information service, 2024, 68(8): 80-96.)
[11] 郑明琪,陈晓慧,刘冰,等.提示学习中思维链生成和增强方法综述[J].计算机科学, 2025, 52(1): 56-64. (ZHENG M Q, CHEN X H, LIU B, et al. Survey of chain-of-thought generation and enhancement methods in prompt learning[J]. Computer science, 2025, 52(1): 56-64.)
[12] SUN Q, HUANG K, YANG X, et al. Consistency guided knowledge retrieval and denoising in LLMs for zero-shot document-level relation triplet extraction[C] //Proceedings of the ACM web conference 2024. New York: Association for computing machinery, 2024: 4407-4416.
[13] 周正达,王昊,汪琳,等.ChatKG：一种基于大语言模型和提示工程的非遗知识图谱构建框架——以中国非遗陶瓷制作工艺为例[J/OL].图书馆杂志, 1-30[2025-11-13]. http://kns.cnki.net/kcms/detail/31.1108.g2.20250224.1136.008.html. (ZHOU Z D, WANG H, WANG L, et al. ChatKG: A framework for constructing intangible cultural heritage knowledge graphs based on large language model and prompt engineering: a case study of Chinese intangible cultural heritage ceramics craft[J/OL]. Library journal, 1-30[2025-05-29]. http://kns.cnki.net/kcms/detail/31.1108.g2.20250224.1136.008.html.)
[14] 刘耀文,夏一雪,张鹏,等.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. (LIU Y W, XIA Y X, ZHANG P, et al. Research on knowledge graph construction and retrieval-enhanced question answering framework for national security intelligence strategy[J]. Journal of intelligence, 2025, 44(7): 165-173.)
[15] 沈振东.科学问题问题域、应答域的难度系数集消解途径研究[J].宁德师范学院学报(哲学社会科学版), 2021(2): 79-85. (SHEN Z D. Research on the resolution of difficulty coefficient set in the problem domain and response domain of scientific problems[J]. Journal of Ningde Normal University(philosophy and social sciences), 2021(2): 79-85.)
[16] 索传军,牌艳欣.科学问题谱系构建研究[J].中国图书馆学报, 2025, 51(1): 61-81. (SUO C J, PAI Y X. The construction of an academic genealogy for scientific problems[J]. Journal of library science in China, 2025, 51(1): 61-81.)
[17] NEWMAN I, COVRIG D M. Building consistency between title, problem statement, purpose, & research questions to improve the quality of research plans and reports[J]. New horizons in adult education and human resource development, 2013, 25(1): 70-79.
[18] 索传军,李木子.我国学术论文研究问题探析——基于2015-2020年图情领域CSSCI发表论文的实证研究[J].图书情报工作, 2021, 65(19): 105-116. (SUO C J, LI M Z. Analysis on the research questions of academic papers in China: empirical research based on papers published in CSSCI in the field of library and information science from 2015 to 2020[J]. Library and information service, 2021, 65(19): 105-116.)
[19] KIM Y, TIAN Y, JEONG Y, et al. Automatic discovery of technology trends from patent text[C] //Proceedings of the 2009 ACM symposium on applied computing. New York: Association for computing machinery, 2009: 1480-1487.
[20] 全国文献工作标准化技术委员会.文摘编写规则： GB6447-86[S/OL].[2025-04-01]. http://c.gb688.cn/bzgk/gb/showGb?type=online&hcno=8713727404B8704955E5CAF11EE3 EB59. (CHINA NATIONAL TECHNICAL COMMITTEE OF STANDARDIZATION FOR DOCUMENTATION. Rules for abstracts and abstracting： GB6447-86[S/OL].[2025-04-01]. http://c.gb688.cn/bzgk/gb/showGb?type=online&hcno=8713727404B8704955E5CAF11EE3EB59.)
[21] 刘冠军.科学问题哲学解与科学解的比较研究[J].南昌大学学报(人文社会科学版), 2001(2): 34-38. (LIU G J. A comparative study of philosophical solution and scientific solution to scientific problems[J]. Journal of Nanchang University(humanities and social sciences), 2001(2): 34-38.)
[22] 刘勘,李冶,石锴文.基于句法和语义关联的科技文献“问题—方法”联合抽取模型[J].知识管理论坛, 2024, 9(4): 353-366. (LIU K, LI Y, SHI K W. “Problem-method” joint extraction model in scientific literature based on syntax and semantic association[J]. Knowledge management forum, 2024, 9(4): 353-366.)
[23] 王忠义,沈雪莹,黄京.科技文献资源中方法知识元的抽取研究[J].情报科学, 2021, 39(1): 13-20. (WANG Z Y, SHEN X Y, HUANG J. Extraction of method knowledge elements in scientific literature[J]. Information science, 2021, 39(1): 13-20.)
[24] 何天文,王红.基于语义语法分析的中文语句困惑度评价[J].计算机应用研究, 2017, 34(12): 3538-3542, 3546. (HE T W, WANG H. Evaluating perplexity of Chinese sentences based on grammar & semantics analysis[J]. Application research of computers, 2017, 34(12): 3538-3542, 3546.)
[25] 中国科学院文献情报中心科技文献知识AI引擎项目组. SciAIEngine中文科研实体识别V3[EB/OL].[2025-09-19]. http://sciengine.las.ac.cn/NKRD_NER_SCI_CN_V3. (SCIAIENGINE GROUP, NATIONAL SCIENCE LIBRARY, CHINESE ACADEMY OF SCIENCES. SciAIEngine Chinese research entity recognition v3[EB/OL].[2025-09-19]. http://sciengine.las.ac.cn/NKRD_NER_SCI_CN_V3.)
[26] OPENAI. Hello gpt-4o[EB/OL].[2025-09-20]. https://openai. com/index/hello-gpt-4o/.
[27] GUO D, YANG D, ZHANG H, et al. DeepSeek-r1: incentivizing reasoning capability in LLMs via reinforcement learning[J/OL]. arXiv, 2025[2025-09-20]. https://arxiv.org/pdf/2501.12948.
[28] YANG A, LI A, YANG B, et al. Qwen3 technical report[J/OL]. arXiv, 2025[2025-09-20]. https://arxiv.org/abs/2505.09388.
Options
文章导航

/

〈 〉