Knowledge-enhanced large language models for automatic lesson plan generation

A well-designed lesson plan is a structured instructional document that helps teachers improve classroom efficiency. Since creating lesson plans is a time-consuming task, recent studies leverage Large Language Models (LLMs) to automatically generate lesson plans inspired by LLMs’ impressive language understanding and generation capability. However, existing LLM-based lesson plan generation approaches still face challenges due to specific real-world educational requirements of lesson plan generation including structural integrity, logical coherence and high accuracy. In this paper, we propose LessonPlanLM, a Knowledge-enhanced Automatic Lesson Plan Generation framework that incorporates LLMs with a lesson plan knowledge base (LPKB). Based on the proposed LPKB, we fine-tuned LLMs to generate standardized and structured lesson plans in a step-by-step manner, addressing core educational requirements. Furthermore, we develop retrieval-augmented fine-tuned (RAFT) LLMs by retrieving relevant documents from the LPKB to support knowledge-aware requirements. To evaluate the performance of generated lesson plans in two-level requirements, we introduce a comprehensive evaluation framework with four distinct perspectives. Experimental results show that our framework can generate more logical and accurate lesson plans. To encourage reproducible research, we make our data and code publicly available at https://github.com/ai4ed/LessonPlan .