Construction and Application of an Obsidian-Based Personal Knowledge Graph and Vector Knowledge Base for Medical Textbooks: A Case Study of Anatomy and Stomatology
Construction and Application of an Obsidian-Based Personal Knowledge Graph and Vector Knowledge Base for Medical Textbooks: A Case Study of Anatomy and Stomatology
Abstract
To address the challenges of extensive content, high terminology density, limited cross-textbook knowledge integration, and insufficient support for self-directed learning in medical education, this study developed a personal knowledge system using the 10th editions of Systematic Anatomy, Regional Anatomy, and Stomatology as source materials. The system integrates Obsidian-based notes for knowledge organization, terminology-based associations for knowledge graph construction, and a vector knowledge base for semantic retrieval. An automated processing pipeline was developed to perform text extraction from scanned PDFs, text normalization, hierarchical heading reconstruction, and chapter-level segmentation for atomic note generation. Terminology hubs and bidirectional links between atomic notes were established to construct a personal knowledge graph, while textbook text and tables were vectorized to support knowledge retrieval. Two deployment architectures were implemented according to the organization of the vector knowledge bases: a single-database architecture and a multi-database architecture. The single-database architecture integrated Systematic Anatomy and Regional Anatomy into a shared vector database, generating 259 atomic notes, 2594 terminology hubs, and 15858 terminology-based bidirectional links. A total of 2512 vector records were indexed, comprising 1350 from Systematic Anatomy and 1162 from Regional Anatomy, with a median retrieval similarity score of 0.708. The multi-database architecture maintained separate vector databases for Systematic Anatomy and Stomatology, generating 442 atomic notes, 2,228 terminology hubs, and 8772 terminology-based bidirectional links. A total of 3670 vector records were indexed, comprising 1709 from Systematic Anatomy and 1961 from Stomatology, with a median retrieval similarity score of 0.736. Terminology-based automatic routing was implemented to direct queries to the corresponding vector databases. Within Obsidian, atomic notes were used to organize and present textbook knowledge; the Graph View visualized the bidirectional-link knowledge network among terminology hubs and external learning resources, and the Dataview plugin dynamically aggregated related atomic and supplementary notes within each terminology hub. The obsidian-opencode plugin, integrated with a local Model Context Protocol (MCP) retrieval service, enabled retrieval-augmented question answering and automatic question generation grounded in textbook content, with source references traceable at three levels: textbook, chapter, and atomic note. The results demonstrate the feasibility of both deployment architectures in supporting source traceability, system portability, and textbook-oriented semantic retrieval. This framework provides a reproducible and low-cost approach to knowledge organization and personalized learning in undergraduate medical education, particularly in anatomy.
关键词
医学教材/Obsidian 笔记/个人知识图谱/个人知识库/检索增强生成/解剖学教学
Key words
medical textbooks/ Obsidian/ personal knowledge graph/ personal knowledge base/ retrieval-augmented generation/ anatomy education
