Aspect-Based Sentiment Analysis of Patient Feedback Using Large Language Models

Abstract

Online medical forums have emerged as vital platforms for patients to share their experiences and seek advice, providing a valuable, cost-effective source of feedback for medical service management. This feedback not only measures patient satisfaction and improves health service quality but also offers crucial insights into the effectiveness of medical treatments, pain management strategies, and alternative therapies. This study systematically identifies and categorizes key aspects of patient experiences, emphasizing both positive and negative sentiments expressed in their narratives. We collected a dataset of approximately 15,000 entries from various sections of the widely used medical forum, patient.info. Our innovative approach integrates content analysis with aspect-based sentiment analysis, deep learning techniques, and a large language model (LLM) to analyze these data. Our methodology is designed to uncover a wide range of aspect types reflected in patient feedback. The analysis revealed seven distinct aspect types prevalent in the feedback, demonstrating that deep learning models can effectively predict these aspect types and their corresponding sentiment values. Notably, the LLM with few-shot learning outperformed other models. Our findings enhance the understanding of patient experiences in online forums and underscore the utility of advanced analytical techniques in extracting meaningful insights from unstructured patient feedback, offering valuable implications for healthcare providers and medical service management.