TextBook-Refiner: RAG-Enhanced Knowledge Graph Construction and Teaching Assistant for Educational Scenarios

TextBook-Refiner is a knowledge integration agent for educational scenarios. By automatically constructing textbook knowledge graphs and integrating RAG technology, it provides teachers with an intelligent teaching assistance workflow, enabling the smart transformation from textbook content to teaching applications. Its core is to enhance teachers' capabilities rather than replace them.

In the digital transformation of education, massive textbook content is difficult to convert into structured, computable knowledge and integrate into teachers' daily workflows. Current tools have polarization issues: basic tools have simple functions, while complex platforms have high learning costs and are disconnected from textbooks. Knowledge graphs (structured knowledge representation) and RAG (avoiding model hallucinations) technologies provide a path to solve this dilemma.

The project is positioned as an intelligent assistant for teachers' workflows, with core capabilities including: 1. Automatic construction of textbook knowledge graphs: identifying knowledge points and relationships, converting them into a network structure; 2. RAG-enhanced teaching assistance: generating syllabi, practice questions, etc., based on the graph, with content strictly derived from textbooks.

Knowledge graph construction pipeline: Document parsing (processing multi-format textbooks, including OCR and layout analysis) → Entity relationship extraction (fine-tuning large models in the education field to identify concepts and relationships) → Graph storage (graph database + vector embedding). The RAG workflow uses hybrid retrieval: structured retrieval from knowledge graphs + semantic retrieval from vector databases, balancing precision and recall.

Supports multiple workflows: lesson preparation assistance (generating knowledge point lists, key and difficult point analysis), differentiated teaching (generating content of different difficulty levels), knowledge association discovery (cross-chapter/subject association queries), and student Q&A preparation (predicting questions and key points for answers).

Addressing high requirements of educational scenarios: Accuracy assurance (RAG limits knowledge sources + consistency checks), interpretability design (displaying knowledge source paths), content safety filtering (multi-level review mechanism).

Limitations: Mainly supports STEM subjects and relies on the standardization of textbook formats. Future directions: Expand textbook format parsing, introduce multi-modal processing, develop student-side applications, and establish a cross-school knowledge graph sharing mechanism.

TextBook-Refiner integrates technology into teachers' workflows, freeing teachers from tedious organizing work, allowing them to focus on teaching design and student interaction, embodying the concept of 'invisible technology, visible education'.