Exploring Data-Driven Care Coaching Generation Using Multimodal Large Language Models for Dementia Care
Abstract
We describe a method for generating data-driven care coaching using multimodal large language models (MLLM) in dementia care. Improving the quality of dementia care requires instruction from experienced caregivers, but opportunities for instruction are limited. Recently, large language models (LLMs) have attracted attention, and we aim to apply them to care coaching to develop a more efficient and effective learning support environment. In this study, we propose a framework for generating care coaching by providing dementia care practice data as input to multimodal LLMs. We have generated care coaching using multimodal LLMs for care practice videos and compared them with conventional human coaching. The results have shown that the data-driven approach using multimodal LLMs is effective in learning dementia care skills.
Atsushi Omata
Research Associate
