Hybrid sentiment analysis in healthcare: Multi-method approach

Sentiment analysis in healthcare discourse is vital for enhancing patient-provider communication, quality of care, and public health responsiveness. Current methods struggle with the nuanced emotional complexities inherent in healthcare narratives. This study proposes a hybrid sentiment analysis framework combining rule-based approaches, a fine-tuned RoBERTa model, and large language models LLMs, optimized for classifying six emotional dimensions: confidence, satisfaction, anxiety, anger, hope, and trust. Evaluated on over 100,000 Reddit posts from health-related subreddits, the hybrid model significantly outperformed individual methods, achieving a macro F1-score of 0.9348 compared to RoBERTa (0.9056) and LLM-based approaches (0.7532). This approach balances interpretability and semantic depth, enabling nuanced, real-time emotion tracking and early detection of shifts in public sentiment. Our findings offer critical implications for improving patient-centered care, clinical decision support, and ethical AI deployment in healthcare contexts.