Abstract: This paper aims to investigate the application of the Large Language Model in educational question-answering systems, and explore its optimization strategies in the educational domain. In recent years, methods based on pre-trained models have garnered much attention in the field of natural language processing. Large Language Model, as a pre-trained language generation model, holds promising potential for reducing development costs and enhancing accuracy in educational question-answering systems. A comprehensive analysis is made from three key aspects: the practical application of Large Language Model in educational question-answering systems, its impact on the educational sector, and optimization approaches. Regarding the practical application in education, the effects of multi-turn question-answering, zero-shot (few-shot) learning, and multi-modal query handling are investigated, and a quantitative analysis is conducted. Additionally, a strategy based on Hard Prompts is explored, aiming at elevating the performance and applicability of Large Language Model in educational question-answering systems. Through a comprehensive evaluation of its strengths and limitations, reference and guidance are provided for intelligent tutoring within the realm of education.

Key words: Large Language Model, educational question-answering system, multi-turn question-answering, zero-shot learning, multi-modal