Presentation Information
[4Yin-B-35]Development and Preliminary Evaluation of a Breast Cancer Peer Support AI Based on Narrative RAG and Persona Adaptation
〇RUEIYANG LIN1, Shuntaro Yada1 (1. University of Tsukuba)
Keywords:
Peer Support,Medical Journey Blog,Personality Traits,Large Language Model (LLM),Chatbots
Peer support alleviates psychological distress in breast cancer patients, yet access is limited by temporal and privacy constraints. We propose a dialogue system combining Retrieval-Augmented Generation (RAG) using real patient narratives and persona adaptation based on Big Five personality traits. The system extracts user treatment data and personality, retrieving similar experiences from a patient blog corpus to generate empathetic responses via persona-matched prompts. With LLM-as-a-Judge, we evaluated the system responses in four modes using a 2 x 2 experimental design (i.e. toggling RAG and persona features), against synthetic patient queries based on Shizuoka Classification. Results showed that RAG significantly improves the peer supportive response quality. Notably, combining RAG with persona adaptation further improved the results, suggesting that integrating narratives with personality-aware dialogue design would enhance AI-driven peer support.