Hybrid Sentiment Analysis of Drug Reviews Using Ml and Lexicon-Based Methods
Kallol Kanti Mondal, Yasmin Akter Bipasha, S A Sabbirul Mohosin Naim, Priya Podder · 2025
The increasing availability of patient-generated drug reviews on digital platforms presents a unique opportunity to assess medication efficacy, adverse effects, and user sentiment comprehensively. However, challenges such as informal review language and inconsistent rating systems limit the efficacy of conventional sentiment analysis techniques. This research introduces a hybrid sentiment analysis method combining lexicon-based strategies and machine learning algorithms to effectively classify user sentiment regarding drug effectiveness and side effects. Utilizing the Drug Review Dataset from the UCI Machine Learning Repository, this study applies extensive preprocessing techniques, including noise reduction and feature extraction via Bag of Words (BoW) and Term FrequencyInverse Document Frequency (TF-IDF) methods. Comparative analyses of various supervised classifiers-Naïve Bayes, Logistic Regression, Decision Trees, LightGBM, and Passive Aggressive Classifier-demonstrate that the TF-IDF embedding with Passive Aggressive Classifier achieves superior results, yielding an accuracy of 98.21 %, precision of 96.25 %, recall of 95.39 %, and$\mathbf{F 1}$-score of$\mathbf{9 5. 8 2 \%}$. These results underscore the benefit of integrating lexicon-based approaches with advanced machine learning for robust sentiment analysis in healthcare.