This research pioneers a knowledge graph-based retrieval-augmented generation (KG-RAG) framework specifically designed for Japanese medical question answering using smaller, more deployable language models.

• Addresses critical privacy constraints preventing commercial LLM use in Japanese healthcare settings
• Demonstrates how knowledge graphs can enhance small-scale LLMs for specialized medical applications
• Provides a pathway for effective medical QA in non-English languages without requiring massive models
• Offers practical solutions for clinical settings where data privacy is paramount

This work matters because it enables accurate medical information access in Japanese healthcare environments while respecting strict privacy regulations, potentially improving patient care without compromising sensitive data.

Exploring the Role of Knowledge Graph-Based RAG in Japanese Medical Question Answering with Small-Scale LLMs