A comparative approach on classification of medical condition drug reviews using multiple traditional and neural network-based models.

This paper uses national language processing techniques to perform sentiment analysis, classifying user-written drug reviews for pain, depression, and birth control. Six machine learning (ML)-based models are employed for analysis, including neural networks (NNs) like the simple NN and convolutional neural networks (CNNs), as well as more conventional models like Adaptive Boosting (AdaBoost), Random Forest (RF), Decision Tree (DT), and Support Vector Classification (SVC). Tokenization used TF-IDF for traditional models and word embeddings for CNN and a simple NN. Using precision, accuracy, recall, and F1-score, CNN achieved the highest F1-score (0.991), followed by NN (0.985) and RF (0.980).