Abstract: Digital textbooks have evolved from version 1.0 to 5.0. and the robust growth of generative artificial intelligence brings new technological affordances for the transformation of digital textbooks. However, there have been structural issues such as linear knowledge presentation, homogeneous learning support, and lagging evaluation feedback. A conceptual framework for developing AI-powered digital textbooks oriented towards human-machine collaboration and a large model embedded solution based on graph-vector fusion enhancement might well address such issues. This solution is based on a three-layer technical architecture of “multi-dimensional representation-intelligent service-collaborative application”: at the knowledge representation layer, a hybrid representation combining knowledge graphs and semantic vector databases is adopted, and multimodal data fusion in textbooks is achieved through triple extraction and vectorization encoding technologies; at the intelligent service layer, a clustered service system based on large model agents is designed that integrates specialized agents, including course design, learning diagnosis, and dialogue guidance, forming modular service units; at the application and collaboration layer, a bidirectional interaction model of “human-led, intelligence-enhanced” is constructed, and three-dimensional application scenarios are proposed, such as teacher-AI collaborative instructional design and student-AI cognitive partnership interaction. The future-visioned development pathways including building education-specific large models, reshaping the textbook compilation industry, and strengthening ethical regulation for human-machine collaboration can provide references for large language models empowering innovative applications of digital textbooks and promoting high-quality educational development.

Key words: digital textbooks, knowledge graph, large language model, graph-vector fusion