An Opinion-Enriched Explainable Semantic Approach for Patient Feedback Triage to Support Clinical Quality Monitoring and Improvement

Abstract

Patient input, though valuable, is often underused for improving clinic quality and safety. Existing semantic similarity methods do well with lexical and contextual links but often miss important opinion cues like sentiment, stance, and severity, which are key for practical healthcare insights. To fix this, we offer an opinion-focused semantic method for automated patient input sorting and clinic quality checks. The setup combines domain-specific embeddings with sentiment and severity modeling, plus interpretable parts that give clear, clinician-focused explanations. By linking input to quality-of-care areas and focusing on high-severity cases, the system makes sure important issues get attention quickly. Tests on de-identified data, like synthetic MIMIC-based input, HCAHPS open-text answers, and adverse event reports, show a clear 12–18% gain in spotting serious cases compared to normal semantic methods, along with better trust and usability among clinical workers. These results suggest that opinion-focused semantic similarity can turn patient input into useful knowledge for ongoing quality improvement in healthcare.