Classifying Drug Ratings Using User Reviews with Transformer-Based Language Models.

Drug review websites such as Drugs.com provide users' textual reviews and numeric ratings of drugs. These reviews along with the ratings are used for the consumers for choosing a drug. However, the numeric ratings may not always be consistent with text reviews and purely relying on the rating score for finding positive/negative reviews may not be reliable. Automatic classification of user ratings based on textual review can create a more reliable rating for drugs. In this project, we built classification models to classify drug review ratings using textual reviews with traditional machine learning and deep learning models. Traditional machine learning models including Random Forest and Naive Bayesian classifiers were built using TF-IDF features as input. Also, transformer-based neural network models including BERT, Bio_ClinicalBERT, RoBERTa, XLNet, ELECTRA, and ALBERT were built using the raw text as input. Overall, Bio_ClinicalBERT model outperformed the other models with an overall accuracy of 87%. We further identified concepts of the Unified Medical Language System (UMLS) from the postings and analyzed their semantic types stratified by class types. This research demonstrated that transformer-based models can be used to classify drug reviews based solely on textual reviews.