Manuscript received February 18, 2024; revised March 4, 2024; accepted March 11, 2024; published July 15, 2024
Abstract—In recent years, there has been a growing need for tailored educational and support measures for children with developmental disorders in Japan. Approximately 8.8% of students in regular classes require special support, necessitating close collaboration and information sharing among schools, families, and welfare facilities. However, teachers and supporters face challenges such as instructional skills, workload, and parent interactions, which cannot be resolved simply by increasing years of experience. Collaboration with external specialized institutions is an effective measure, but few schools engage in regular information sharing. This study proposes the use of Large Language Models (LLMs) and Retrieval- Augmented Generation (RAG) based on the support information and behavioral history of students accumulated in the Individualized Educational Support System. The proposed approach offers practical solutions to the lack of experience in special education, insufficient collaboration, and workload issues. Experiments with a question-answering system using multiple models confirmed the possibility of providing answers based on content specific to each child. The effectiveness of a summary system in providing individualized support was confirmed through evaluations by actual supporters. The proposed approach demonstrates the potential for addressing the challenges faced in supporting children with developmental disorders in Japan.
Keywords—developmental disabilities, collaborative support, individualized educational support, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG)
Cite: Masaki Kato, Makoto Koshino, Sakiko Ogoshi, and Yasuhiro Ogoshi, "Educational Support System for Children with Developmental Disabilities through Offline Large Language Models and Retrieval-Augmented Generation," International Journal of Information and Education Technology vol. 14, no. 7, pp. 955-960, 2024.