Using Sentiment Analysis & Emotion Recognition in Patient Feedback to Improve Care Quality & Revenue

Amruta Inc's Dr. Beju Rao, PhD, Vishwa Makesh, and coauthors presented a breakout session at Voice and AI 2023, the leading conference for Natural Language and Generative AI.

Patients provide written and spoken feedback on their healthcare interactions. Satisfied patients mean a higher reputation and revenue for health systems. Knowing when and how patients are expressing and conveying negative and positive sentiments and emotions provide actionable insights to hospital administrators and operators.

Natural Language Processing (NLP) and eXplainable AI (XAI) are crucial in processing patient feedback to analyze sentiments and predict emotions. Accurate sentiment analysis and emotion recognition involve understanding the semantics, context, syntax and ever evolving slang and abbreviations. Sentiment analysis and scoring is done after preprocessing the text by expanding the abbreviations, correcting the spelling, and replacing the slang words. Case, punctuation (e.g., exclamation marks and emoticons), and qualifiers (adverbs of accentuation, adverbs of frequency, and negation) impact the sentiment score, among others. Stop words are removed, lowercase conversion is done, and lemmatization is done before we encode the words as input to the Machine Learning (ML) and Deep Learning (DL, Neural Net) emotion prediction models. We design our algorithms to be explainable, using both ante-hoc and post-hoc explainability.

We predict eight emotions, Happiness, Trust, Surprise, Anticipation, Sadness, Anger, Fear, and Disgust, at various intensity levels. Linguists labeled the feedback for these emotions in the training data. We implemented a continuous refinement algorithm to use Linguists input only when predictions are wrong.

Hospital administrators are using the sentiment scores and emotions recognized to intervene across the nurse and other operational staff behaviors as well as to improve the facility conditions to increase the patient satisfaction. The explainable narratives in emotion recognition are invaluable tools for carrying out the interventions and improvements.