Multi-Class Text Classification Using Machine Learning Models for Online Drug Reviews

Abstract

The reviews that are present in different forms on the Internet can provide valuable insights into the opinions of the users that are spread across a wide range of geographical space in the most time and cost-efficient manner. This information can be used to improve the quality or assess the efficiency of a product over any given domain. In this research, analysis of the users' online reviews within the field of pharmaceuticals is presented. These reviews consist primarily of the information regarding the usefulness of drugs or the side effects they have caused. As much as it is important to find a measure of the efficiency of a drug, it is also essential to determine the medical condition for which the drug is manifesting its effects, be it positive or negative. In this research, six different supervised machine learning classifiers are deployed to find the most efficient model to predict the medical condition based on the users' reviews. The classifiers used are as follows: Multinomial Naive Bayes, Multinomial Logistic Regression, Linear Support Vector Classifier (SVC), Decision Trees, Extra Trees, and Random Forests. The results demonstrate that among all the classifiers used, Linear SVC proved to be the most efficient when considering its Precision, Recall, F1score and the time it takes to train and test on the given data.