图书情报工作

图书情报工作 >

2026 , Vol. 70 >Issue 12: 92 - 103

DOI: https://doi.org/10.13266/j.issn.0252-3116.2026.12.008

研究论文

大语言模型驱动的专利技术功效矩阵构建方法研究

  • 周洪 ,
  • 赵展一 ,
  • 魏雪梅 ,
  • 董金鑫
展开
  • 1 中国科学院武汉文献情报中心 武汉 430071;
    2 中国科学院文献情报中心 北京 100190
周洪，副研究员，博士；赵展一，助理研究员，博士；魏雪梅，专利咨询师；董金鑫，助理研究员，博士，通信作者，E-mail：dongjx@mail.whilb.ac.cn。

收稿日期: 2025-04-17

  修回日期: 2025-06-23

  网络出版日期: 2026-06-09

基金资助

本文系中国科学院文献情报领域引进优秀人才计划择优支持项目“材料制造领域多源科学数据的结构化挖掘与智能解析方法研究”（项目编号：E4KZ071）研究成果之一。

收起

Research on the Construction Method of Patent Technology Function Matrix Driven by Large Language Model

  • Zhou Hong ,
  • Zhao Zhanyi ,
  • Wei Xuemei ,
  • Dong Jinxin
Expand
  • 1 National Science Library(Wuhan), Chinese Academy of Sciences, Wuhan 430071;
    2 National Science Library, Chinese Academy of Sciences, Beijing 100190
Zhou Hong, Associate Research Fellow, PhD; Zhao Zhanyi, Assistant Research Fellow, PhD; Wei Xuemei, Patent Consultant; Dong Jinxin, Assistant Research Fellow, PhD, Corresponding Author, E-mail: dongjx@mail.whilb.ac.cn.

Received date: 2025-04-17

  Revised date: 2025-06-23

  Online published: 2026-06-09

Supported by

This work is supported by Chinese Academy of Sciences Talent Introduction Program for Library and Information Science with Selective Support project (Grant No. E4KZ071).

Fold

摘要

[目的/意义] 旨在提出一种基于大语言模型的专利技术功效矩阵智能生成方法，以提升技术功效矩阵构建的效率和精准度，为知识产权智能分析提供支持。[方法/过程] 首先，利用BERTopic模型对专利文本进行深度语义嵌入和非线性降维，动态粒度调节生成层次化技术主题；然后，基于DeepSeek-R1设计两轮问答动态生成机制，提取和优化技术功效短语，形成层次化和结构化的功效维度体系；最后，通过共现分析构建技术功效矩阵，并进行覆盖率、准确性、可靠性检验。[结果/结论] 以人形机器人为例开展实证研究，结果表明，该方法在技术功效覆盖率、准确性和可靠性等核心指标上表现良好，引入动态粒度调整机制和提示工程与层次聚类的协同优化方法，可以为专利技术功效分析提供兼具细粒度解析与趋势洞察能力的支撑。

关键词： 技术功效矩阵; 大语言模型; 专利分析; DeepSeek

本文引用格式

周洪 , 赵展一 , 魏雪梅 , 董金鑫 . 大语言模型驱动的专利技术功效矩阵构建方法研究[J]. 图书情报工作, 2026 , 70(12) : 92 -103 . DOI: 10.13266/j.issn.0252-3116.2026.12.008

Abstract

[Purpose/Significance] This study proposes an intelligent generation method for patent technology function matrices based on large language models. It aims to enhance the efficiency and accuracy of matrix construction and support intelligent intellectual property analysis. [Method/Process] First, the BERTopic model is utilized for deep semantic embedding and nonlinear dimensionality reduction of patent texts, generating hierarchical technology topics with dynamic granularity. Then, a two-round question-and-answer dynamic generation mechanism is designed based on the DeepSeek-R1 model to extract and optimize technology function phrases, forming a hierarchical and structured function dimension system. Finally, a technology function matrix is constructed through co-occurrence analysis and subsequently validated through tests for coverage, accuracy, and reliability. [Result/Conclusion] An empirical study on humanoid robots patents demonstrates that the proposed method performs well in core indicators such as technology function coverage, accuracy, and reliability. The introduction of dynamic granularity adjustment mechanisms and the collaborative optimization of prompt engineering and hierarchical clustering can provide support for intellectual property intelligent analysis, offering fine-grained analysis and trend insight capabilities.

Key words： technology function matrix; large language model; patent analysis; DeepSeek

参考文献

[1] 王凌燕, 方曙, 季培培. 利用专利文献识别新兴技术主题的技术框架研究[J]. 图书情报工作, 2011, 55(18): 74-78, 23. (WANG L Y, FANG S, JI P P. Using patent documents to study the technology framework of detecting emerging technology topics[J]. Library and information service, 2011, 55(18): 74-78, 23.)
[2] HUANG J Y, HSU H T. Technology-function matrix based network analysis of cloud computing[J]. Scientometrics, 2017, 113: 17-44.
[3] 王学昭, 赵萍, 赵亚娟, 等. “技术-功效”视角下的专利布局形势揭示与风险判定[J]. 图书情报工作, 2021, 65(16): 73-80. (WANG X Z, ZHAO P, ZHAO Y J, et al. The identification of patent layout situation and risk based on technology-effect matrix[J]. Library and information service, 2021, 65(16): 73-80.)
[4] 刘春江, 李姝影, 刘自强, 等. 面向多维技术功效分析的专利技术功效矩阵构建方法研究[J]. 情报理论与实践, 2023, 46(12): 167-174. (LIU C J, LI S Y, LIU Z Q, et al. Research on the construction method of patent technology / effect matrix for multidimensional patent technology / effect analysis[J]. Information studies: theory & application, 2023, 46(12): 167-174.)
[5] 陈颖, 张晓林. 专利技术功效矩阵构建研究进展[J]. 现代图书情报技术, 2011(11): 1-8. (CHEN Y, ZHANG X L. Research progress on construction of patent technology-effect matrix[J]. Data analysis and knowledge discovery, 2011(11): 1-8.)
[6] HE C, TAN R, PENG S, et al. Research on technology-function matrix construction for patent layout[J]. International journal of systematic innovation, 2020, 6(2): 53-62.
[7] TRAPP S, WARSCHAT J. LLM-based extraction of contradictions from patents[C]//International TRIZ future conference. Cham: Springer Nature, 2024: 3-19.
[8] LEE S, LEE J, PARK S, et al. Large language model-based R&D solution analysis approach using problem-solution information of patents[J]. Journal of intellectual property, 2024, 19(3): 155-180.
[9] WANG H Y, CHANG S H, CHUANG C Y. Integrating generative artificial intelligence techniques into technology function matrix analysis[J]. World patent information, 2025, 81: 102352.
[10] 赵建飞, 陈挺, 王小梅, 等. 基于大语言模型知识自蒸馏的无标注专利关键信息抽取[J]. 数据分析与知识发现, 2024, 8(Z1): 133-143. (ZHAO J F, CHEN T, WANG X M, et al. Extracting key information from unlabeled patents based on knowledge self-distillation of large language model[J]. Data analysis and knowledge discovery, 2024, 8(Z1): 133-143.)
[11] RYU S, LEE S. Development of a technology tree using patent information[J]. Advanced engineering informatics, 2024, 59: 102277.
[12] 胡明昊, 田昌海, 易子博, 等. 基于科技信息挖掘的大规模深度技术树构建[J]. 情报理论与实践, 2024, 47(4): 10-15. (HU M H, TIAN C H, YI Z B, et al. Large-scale deep technology tree construction based on technological information mining[J]. Information studies: theory & application, 2024, 47(4): 10-15.)
[13] 陈颖, 张晓林. 专利中技术词和功效词识别方法研究[J]. 现代图书情报技术, 2011(12): 24-30. (CHEN Y, ZHANG X L. Study on the differentiating method of Technical and Effect Words in Patent[J]. New technology of library and information service, 2011(12): 24-30.)
[14] 翟东升, 张京先, 胡等金. 基于SAO结构和词向量的专利技术功效图自动构建研究[J]. 情报理论与实践, 2020, 43(3): 116-123. (ZHAI D S, ZHANG J X, HU D J, et al. Research on automatic construction of patent technology-effect map based on SAO structure and word embedding[J]. Information studies: theory & application, 2020, 43(3): 116-123.)
[15] 陈颖, 张晓林.专利技术功效矩阵构建词汇模型研究[J]. 情报科学, 2012, 30(11): 1704-1708, 1719. (CHEN Y, ZHANG X L. Research on lexical model of construction of patent technology-effect matrix[J]. Information science, 2012, 30(11): 1704-1708, 1719.)
[16] 余梦霞, 张宇娥, 凌世婷, 等. 专利技术功效视域下的领域重点研发方向布局研究——以第三代功率半导体领域为例[J]. 图书情报工作, 2022, 66(17): 116-128. (YU M X, ZHANG Y E, LING S T. Study on the layout of the field’s key research directions from the perspective of patent technology-effect: taking the third-generation power semiconductor field as an example[J]. Library and information service, 2022, 66(17): 116-128.)
[17] 向姝璇, 李睿.专利技术功效特征自动抽取方法探索——以6G领域为例[J]. 中国发明与专利, 2021, 18(4): 3-9. (XIANG Z X, LI R. An improved technology-function features extraction method of patents: an case study of 6G domain[J]. China invention & patent, 2021, 18(4): 3-9.)
[18] 刘鹏, 闫煜析, 冯立杰, 等. 用户需求导向下基于三级技术功效矩阵的产品创新机会识别[J]. 情报理论与实践, 2023, 46(8): 138-146. (LIU P, YAN Y X, FENG L J, et al. Product innovation opportunity identification based on three-level technology-effect matrix under user demand orientation[J]. Information studies: theory & application, 2023, 46(8): 138-146.)
[19] 张金柱, 于文倩, 李溢峰. 利用技术功效语义关联构建技术实现路径[J]. 图书馆论坛, 2021, 41(3): 31-41. (ZHANG J Z, YU W Q, LI Y F. Technology implementation path based on technology effect semantic correlation[J]. Library tribune, 2021, 41(3): 31-41.)
[20] 王丽然.专利技术词与功效词抽取系统的设计与实现[D]. 北京:首都经济贸易大学, 2022. (WANG L R. Design and implementation of patented technical word and effect word extraction system[D]. Beijing: Capital University of Economics and Business, 2022.)
[21] 白如江, 陈启明, 张玉洁, 等. 基于ChatGPT+Prompt的专利技术功效实体自动生成研究[J]. 数据分析与知识发现, 2024, 8(4): 14-25. (BAI R J, CHEN Q M, ZHANG Y J, et al. Generating effectiveness entities of patent technology based on ChatGPT+Prompt[J]. Data analysis and knowledge discovery, 2024, 8(4): 14-25.)
[22] 王奎芳, 吕璐成, 孙文君, 等. 基于大模型知识蒸馏的专利技术功效词自动抽取方法研究：以车联网V2X领域为例[J]. 数据分析与知识发现, 2024, 8(Z1): 144-156. (WANG K F, LV L C, SUN W J, et al. Automatic extraction of patent technical effect words based on large model knowledge distillation: A case study of v2x in internet of vehicle[J]. Data analysis and knowledge discovery, 2024, 8(Z1): 144-156.)
Options
文章导航

/

〈 〉