Sentiment Analysis and Emotional Language as Predictors of Drug Satisfaction in User Reviews

Abstract

This study investigates how emotional expressions in user-generated drug reviews predict satisfaction ratings using sentiment analysis and emotion detection. By analyzing over 370,000 reviews from the UCI Machine Learning Repository, the study aims to bridge gaps in understanding the emotional drivers behind user satisfaction across different drug categories. For sentiment analysis, VADER, a Python-based lexicon tool, was used to categorize sentiment polarity, while the NRC Word-Emotion Lexicon provided a nuanced mapping of emotions like joy, sadness, and anger. Results reveal that emotions such as joy and trust are positively correlated with higher ratings, while anger and disgust are linked to lower satisfaction. However, the R-squared value (~0.043) indicates that emotions alone do not fully predict ratings, highlighting the need to consider additional factors like drug efficacy and side effects. This low R-squared value suggests that while emotions significantly influence satisfaction, other elements play a substantial role. The study's findings have critical implications for pharmaceutical companies and healthcare providers, suggesting the need for emotion-driven marketing strategies and improved patient support systems. Future research could explore more advanced machine learning models, such as BERT or GPT-based approaches, and investigate specific user demographics or drug side effects to enhance predictive accuracy.