Controlled Generation of Educational Content Based on Knowledge-Oriented Models

Abstract

This paper proposes a method for controlled educational content generation based on knowledge-oriented models and Large Language Models (LLMs). The method involves constructing a semantic context according to a specified learning objective, identifying mandatory concepts, semantic relationships, and pedagogical constraints, and using them to guide the generation process. To ensure the quality of the generated content, an evaluation framework is proposed based on semantic coverage, factual correctness, pedagogical alignment, difficulty compliance, and content completeness. An iterative “generation → validation → evaluation → feedback → regeneration” mechanism is implemented to automatically refine the generated material when it fails to satisfy the predefined requirements. The proposed approach provides a foundation for improving the controllability and quality of automatically generated educational materials and facilitates their subsequent integration into intelligent and adaptive learning environments.