基于ChatGPT API和提示词工程的专利知识图谱构建  

Construction of Patent Knowledge Graphs Based on ChatGPT API and Prompt Engineering

作　　者：张玲玲 黄务兰 Zhang Lingling;Huang Wulan(School of Business Information,Shanghai Business School,Shanghai 200235)

机构地区：[1]上海商学院商务信息学院,上海200235

出　　处：《情报杂志》2025年第3期180-187,共8页Journal of Intelligence

基　　金：国家社会科学基金一般项目“面向智库建设的图书馆知识服务模式和创新路径研究”(编号:18BTQ058)研究成果。

摘　　要：[研究目的]在信息爆炸的时代背景下,专利数据的快速增长为知识管理和分析带来了新的挑战。该文旨在探讨利用ChatGPT从专利摘要中抽取信息,构建专利知识图谱,以提升知识管理和分析的效率和准确性。[研究方法]从中国知网专利数据库选取了智能驾驶领域的专利摘要,利用ChatGPT进行信息抽取。为实现高效批量处理,采用了ChatGPT API接口与模型进行交互。为确保信息抽取的准确性,多次迭代和优化提示词,设计了系统消息、助手消息及用户消息三种角色,通过模拟对话场景,引导模型精确抽取实体与关系。[研究结果/结论]研究结果表明,ChatGPT成功从1126份专利摘要中提取了丰富的五元组信息,并以此为基础构建了专利知识图谱。与传统方法如Bert2Keras相比,ChatGPT在精确率、召回率及F1值等关键指标上均表现出明显优势,分别达到了88.2%、88.3%和88.3%,远超Bert2Keras的34.7%、9%和14.6%。最后,利用抽取的实体关系和Neo4j技术,成功地构建了知识图谱并完成了可视化展示,便于通过Cypher语句进行查询操作。该研究不仅证实了ChatGPT在专利知识图谱构建中的可行性,也为其在知识产权管理、技术研发及竞争情报分析等方面的智能化应用奠定了基础。[Research purpose]In the context of the information explosion era,the rapid growth of patent data presents new challenges for knowledge management and analysis.This study aims to explore the use of ChatGPT for information extraction from patent abstracts to construct patent knowledge graphs,thereby enhancing the efficiency and accuracy of knowledge management and analysis.[Research method]Patent abstracts in the field of intelligent driving were selected from the China National Knowledge Infrastructure(CNKI)patent database,and ChatGPT was utilized for information extraction.To achieve efficient batch processing,the ChatGPT API was used for interaction.To ensure the accuracy of information extraction,multiple iterations and optimizations of prompts were performed,designing three roles:system messages,assistant messages,and user messages.These roles guided the model to accurately extract entities and relationships through simulated dialogue scenarios.[Research result/conclusion]The results indicate that ChatGPT successfully extracted rich quintuple information from 1126 patent abstracts and constructed a patent knowledge graph based on this information.Compared with traditional methods such as Bert2Keras,ChatGPT demonstrated significant advantages in key metrics such as precision,recall and F1-score,achieving 88.2%,88.3%and 88.3%,respectively,far surpassing Bert2Keras's 34.7%,9%and 14.6%.Finally,using the extracted entities,relationships and Neo4j technology,a knowledge graph was successfully constructed and visualized,facilitating query operations through Cypher statements.This study not only confirms the feasibility of using ChatGPT in patent knowledge graph construction but also lays the foundation for its intelligent application in intellectual property management,technological development,and competitive intelligence analysis.

关 键 词：ChatGPT API 提示词工程 专利知识图谱 实体关系抽取 智能驾驶 

分 类 号：G251[文化科学—图书馆学]