Demographic Prediction Based on User Reviews about Medications

Abstract

Drug reactions can be extracted fromuser reviews provided on the Web, and processingthis information in an automated way represents anovel and exciting approach to personalized medicineand wide-scale drug tests. In medical applications,demographic information regarding the authors ofthese reviews such as age and gender is of primaryimportance; however, existing studies usually assumethat this information is available or overlook theissue entirely. In this work, we propose andcompare several approaches to automated mining ofdemographic information from user-generated texts.We compare modern natural language processingtechniques, including feature rich classifiers, extensionsof topic models, and deep neural networks (bothconvolutional and recurrent architectures) for thisproblem.