LANG: A Lesson Plan Generation Framework via Multi-Form Interaction with Large Language Models

Lesson plans are essential tools for structuring and organizing the teaching process. However, the traditional approach to creating lesson plans demands considerable time and effort from teachers, as they must review extensive literature and collate relevant information. Thus, developing a technology capable of automatically generating lesson plans holds great significance. Such technology not only alleviates teachers' workloads but also enhances the efficiency of lesson preparation. This paper addresses this need by leveraging the capabilities of large language models (LLMs) and improving the quality of generated lesson plans through various input forms, enhanced generation methods, and teacher-interaction mechanisms. In particular, we present the ''Plan10k'' dataset-a high-quality, bilingual collection of lesson plans. Based on this dataset, we propose the LANG framework, which comprises three key modules: a query rewriting module, a lesson plan generation module, and a chapter correction module. The query rewriting module processes diverse teacher inputs, including specific knowledge points and textbooks, ensuring contextual accuracy. The lesson plan generation module then constructs comprehensive lesson plans, while the chapter correction module integrates retrieval tools to rectify errors and improve output quality. Importantly, the framework supports multiple forms of interaction with intermediate results, offering teachers flexibility and control over the generation process. Extensive experiments validate the effectiveness of our framework, demonstrating significant improvements in performance and high teacher satisfaction. All codes and datasets are publicly available at https://github.com/ssakana/LANG.