针对当前学术知识服务中关键科研要素信息不够全面详尽的问题，提出一个两库三维的知识库框架，并采用机器学习结合大语言模型方法，对学术论文中摘要、对比型表格和对比型图片等多模态数据进行科研要素抽取，进而融合相关结果并附加同义词表和提示模板库建设科研要素知识库。本文以《数据分析与知识发现》期刊2019-2023年部分数据为例，进行示范性知识库建设，并开发可视化问答系统界面，进行实证研究，结果对比证明该知识库框架和构建方法在学术知识服务方面的优势。

To address the issue of insufficient detail about the crucial scientific research elements in academic knowledge services, a knowledge base framework with two bases and three aspects is proposed. We also develop a method combining machine learning with large language models to extract scientific research elements in multimodal academic papers data, such as abstracts, comparative tables, and comparative images, and then integrate them with thesaurus and prompt template library to construct a scientific research elements knowledge base. Using data from the journal Data Analysis and Knowledge Discovery from 2019 to 2023 as an example, the construction of the knowledge base demo and development of a visual question-answering system were conducted as an empirical study, which demonstrates the advantages of this knowledge base framework and construction method in knowledge services.