Predicting online reviewer popularity: A comparative analysis of machine learning techniques

Abstract

Online customer reviews have been found to vary in their level of influence on customers’ purchase decisions depending on both review and reviewer characteristics. It is logical to expect reviews written by popular reviewers to wield more influence over customers, and therefore an investigation into factors which can help explain and predict reviewer popularity should have high academic and practical implications. We made a novel attempt at using machine learning techniques to classify reviewers into high/low popularity based on their profile characteristics. We compared five different models, and found the neural network model to be the best in terms of overall accuracy (84.2%). Total helpfulness votes received by a reviewer was the top determinant of popularity. Based on this work, businesses can identify potentially influential reviewers to request them for reviews. This research-in-progress can be extended using more factors and models to further enhance the accuracy rate. © Springer International Publishing AG 2017.