A Self-reflection Rewriting Method for Readability of Answers to Question about Spinal Cord Injury

Presentation Time: 10:33 AM - 10:45 AM

Abstract Keywords: Large Language Models (LLMs), Delivering Health Information and Knowledge to the Public, Evaluation
Working Group: Natural Language Processing Working Group
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Retrieval Augmented Generation (RAG) often produces overly complex answers in biomedical Q&A. We evaluated three rewriting methods: (1) target grade-level prompt, (2) basic rewrite prompt, and (3) self-reflection with GPT-4o and Llama 3.3 70B-Instruct models. All lowered Flesch–Kincaid scores, but (1) and (2) still exceeded 8th grade level, while (3) hit the target for GPT-4o but not Llama. Despite lower semantic similarity, self-reflection retained over 93% similarity, indicating effective rewrites with minimal information loss.

Speaker(s):
Bayu Aryoyudanta, Master
University of Pittsburgh

Author(s):
Bayu Aryoyudanta, Master - University of Pittsburgh; Maria Yuliana, Bachelor - University of Pittsburgh; I Made Agus Setiawan, PhD - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Brad E. Dicianno, MD - University of Pittsburgh; Yong Kyung Choi, PhD, MPH - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh;