Automated Question Bank Generation Driven by Large Language Models： Cross-Modal Parsing and Knowledge Point Alignment for PDF Documents

SI Bin

doi:10.11907/rjdk.251489

您当前的位置：

首页 >

文章列表页 >

Automated Question Bank Generation Driven by Large Language Models： Cross-Modal Parsing and Knowledge Point Alignment for PDF Documents

Artificial Intelligence | 更新时间：2026-04-30

- Automated Question Bank Generation Driven by Large Language Models： Cross-Modal Parsing and Knowledge Point Alignment for PDF Documents
- Software Guide Vol. 25, Issue 4, Pages: 104-109(2026)
- 作者机构：
  
  萍乡市第二人民医院信息科，江西萍乡 337000
- 作者简介：
- 基金信息：
- DOI：10.11907/rjdk.251489
  CLC： TP182
- Received：27 July 2025，
  
  Published：15 April 2026
- 稿件说明：
移动端阅览
司彬.大语言模型驱动的题库自动生成：面向PDF的跨模态解析与知识点对齐[J].软件导刊,2026,25(04):104-109.

SI Bin.Automated Question Bank Generation Driven by Large Language Models： Cross-Modal Parsing and Knowledge Point Alignment for PDF Documents[J].Software Guide,2026,25(04):104-109.
司彬.大语言模型驱动的题库自动生成：面向PDF的跨模态解析与知识点对齐[J].软件导刊,2026,25(04):104-109. DOI： 10.11907/rjdk.251489.

SI Bin.Automated Question Bank Generation Driven by Large Language Models： Cross-Modal Parsing and Knowledge Point Alignment for PDF Documents[J].Software Guide,2026,25(04):104-109. DOI： 10.11907/rjdk.251489.

摘要

随着在线教育的快速发展，高质量题库的自动化生成需求日益迫切。现有题库自动生成方法主要分为两类：基于教材内容生成的方法存在准确率不足的问题，而基于PDF题库解析的方法由于格式多样性导致普适性较差。为此，提出一种结合知识图谱与混合Embedding的题库生成系统。采用多模态PDF解析技术处理文本、公式等异构内容；构建知识点图谱，实现题目—答案—知识点的语义关联；设计混合匹配算法，融合结构特征与语义特征进行精准对齐。实验结果表明，所提系统在问题—答案解析匹配任务中达到70.68%的准确率，在题目—知识点对齐任务中达到82.95%的准确率。该系统具有良好的普适性，与教师审核机制结合后可快速应用于CFA考试、医师资格考试等专业领域。

Abstract

With the rapid development of online education， the automatic generation of high-quality question banks is increasingly urgent. The existing methods for automatic generation of question bank can be divided into two categories： the method based on textbook content has insufficient accuracy， while the method based on PDF question bank analysis has poor universality due to the diversity of formats. Therefore， a question bank generation system combining knowledge map and hybrid embedding is proposed， which uses multimodal PDF parsing technology to process text， formula and other heterogeneous content； Construct a map of knowledge points to realize the semantic association of question answer knowledge points； A hybrid matching algorithm is designed to fuse structural features and semantic features for precise alignment. The experimental results show that the proposed system achieves an accuracy rate of 70.68% in the question answer analysis matching task and 82.95% in the question knowledge point alignment task. The system has good universality and can be quickly applied to professional fields such as CFA examination， physician qualification examination， etc. after being combined with the teacher review mechanism.

关键词

Keywords

references

FLOR M ， RIORDAN B . A semantic role-based approach to open-domain automatic question generation ［C］// Proceedings of the Thirteenth Workshop on Innovative Use of NLP for Building Educational Applications ， 2018 ： 254 - 263 . doi: 10.18653/v1/w18-0530 http://dx.doi.org/10.18653/v1/w18-0530

RAO P R ， JHAWAR T N ， KACHAVE Y A ， et al . Generating QA from rule-based algorithms ［C］// International Conference on Electronics and Renewable Systems （ICEARS）， 2022 ： 1697 - 1703 . doi: 10.1109/icears53579.2022.9751723 http://dx.doi.org/10.1109/icears53579.2022.9751723

SRIVASTAVA A ， SHINDE S ， PATEL N ， et al . Questionator-automated question generation using deep learning ［C］// International Conference on Emerging Trends in Information Technology and Engineering （ic-ETITE）， 2020 ： 1 - 5 . doi: 10.1109/ic-etite47903.2020.212 http://dx.doi.org/10.1109/ic-etite47903.2020.212

GABAJIWALA E ， MEHTA P ， SINGH R ， et al . Quiz maker：automatic quiz generation from text using NLP ［C］// Singapore：Proceedings of 4th International Conference on Futuristic Trends in Network and Communication Technologies ， 2022 ： 523 - 533 . doi: 10.1007/978-981-19-5037-7_37 http://dx.doi.org/10.1007/978-981-19-5037-7_37

LI D ， LI Z ， YANG Y ， et al . Knowledge graph-enhanced large language model for domain-specific question answering systems［DB/OL］ . https：//www.techrxiv.org/inst/26407 https://www.techrxiv.org/inst/26407 . doi: 10.36227/techrxiv.172963127.77889256/v1 http://dx.doi.org/10.36227/techrxiv.172963127.77889256/v1

LIU L ， ZHOU Y ， MA J ， et al . Domain-specific question-answering systems： a case study of a carbon neutrality knowledge base ［J］. Sustainability ， 2025 ， 17 （ 5 ）： 2192 . doi: 10.3390/su17052192 http://dx.doi.org/10.3390/su17052192

LIU Y ， ZHANG H ， WANG L ， et al . DeepLayout： a multi-modal document layout analysis framework for complex PDFs ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence ， 2022 ， 44 （ 8 ）： 4123 - 4136 .

CHEN X ， SMITH A ， BROWN B . PDF-TREX： an automated pipeline for table extraction and reconstruction from PDF documents ［J］. ACM Transactions on Document Systems ， 2021 ， 15 （ 3 ）： 1 - 24 .

SMITH A ， JOHNSON B ， LEE C . A comprehensive survey on PDF document parsing： from layout analysis to knowledge extraction ［J］. ACM Computing Surveys ， 2023 ， 55 （ 4 ）： 1 - 36 .

CHEN R ， WANG P ， LI S ， et al . ChartSense： automated data extraction from PDF charts via deep layout segmentation ［J］. IEEE Transactions on Visualization and Computer Graphics ， 2023 ， 29 （ 6 ）： 2875 - 2888 .

LIU Y ， CHEN Q ， WU X F . Deep learning-based table recognition in PDFs： a survey ［J］. Acta Automatica Sinica ， 2022 ， 48 （ 8 ）： 1845 - 1862 .

刘洋，陈强，吴晓峰 . 基于深度学习的PDF表格识别技术综述［J］. 自动化学报， 2022 ， 48 （ 8 ）： 1845 - 1862 .

LI H ， ZHANG L ， LIU Y . Joint extraction of cross-modal text and formulas in academic PDFs ［J］. Journal of Software ， 2022 ， 33 （ 5 ）： 1789 - 1804 .

李航，张磊，刘洋 . 学术PDF中跨模态文本与公式的联合提取方法［J］. 软件学报， 2022 ， 33 （ 5 ）： 1789 - 1804 .

LI H ， WANG W ， ZHANG Q . Chinese textbook question answering system based on knowledge graph ［J］. Journal of Chinese Information Processing ， 2022 ， 36 （ 4 ）： 78 - 89 .

李华，王伟，张强 . 基于知识图谱的中文教材问答系统构建［J］. 中文信息学报， 2022 ， 36 （ 4 ）： 78 - 89 .

WANG X ， GAO T ， ZHU Z ， et al . KEPLER： a unified model for knowledge embedding and pre-trained language representation ［C］// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing ， 2020 ： 4175 - 4186 . doi: 10.1162/tacl_a_00360 http://dx.doi.org/10.1162/tacl_a_00360

LAMPLE G ， BALLESTEROS M ， SUBRAMANIAN S ， et al . Neural architectures for named entity recognition ［C］// Proceedings of NAACL-HLT ， 2016 ： 260 - 270 . doi: 10.18653/v1/n16-1030 http://dx.doi.org/10.18653/v1/n16-1030

DEVLIN J ， CHANG M W ， LEE K ， et al . BERT： pre-training of deep bidirectional transformers for language understanding ［C］// Proceedings of NAACL-HLT ， 2019 ： 4171 - 4186 . doi: 10.18653/v1/n18-2 http://dx.doi.org/10.18653/v1/n18-2

LEWIS P ， PEREZ E ， PIKULIK R ， et al . Retrieval-augmented generation for knowledge-intensive NLP tasks ［J］. Advances in Neural Information Processing Systems ， 2020 ， 33 ： 9459 - 9474 .. doi: 10.48550/arXiv.2005.11401 http://dx.doi.org/10.48550/arXiv.2005.11401

XIONG W ， HOANG T ， WANG W Y . DeepPath： a reinforcement learning method for knowledge graph reasoning ［C］// Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing ， 2017 ： 564 - 573 .. doi: 10.18653/v1/d17-1060 http://dx.doi.org/10.18653/v1/d17-1060

DIFY Team . Dify： an open-source LLM application development platform ［EB/OL］. https：//docs.dify.ai/ https://docs.dify.ai/ .

DIFY Team . Dify community edition documentation ［EB/OL］. https：//docs.dify.ai/community-edition https://docs.dify.ai/community-edition .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Research on the Application of AI Agents in Blended Learning

Information Extraction in Materials Science Based on Large Language Model

Knowledge Graph-Based Curriculum Resource Integration and Higher Education Teaching Reform

Intelligent Tutoring System for Software Testing Driven by Large Language Models and Knowledge Graphs

Innovative Design of Blended Teaching for Basic Computer Science Course Based on Knowledge Graph

Related Author

YUAN En

CHEN Weiwei

TANG Yanqin

WU Yongfen

LIU Ling

ZHANG Wenyu

CHEN Han

SHI Zongbin

Related Institution

Command & Control Engineering College， Army Engineering University of PLA

National Science Library， Chinese Academy of Sciences

School of Economics and Management， University of Chinese Academy of Sciences

Soochow College，Soochow University

School of Software， Northwestern Polytechnical University

⁰