Predicting the Usefulness of E-Commerce Products' Reviews Using Machine Learning Techniques

User-generated reviews are an essential component of e-commerce platforms. The presence of a large number of these reviews creates an information overload problem, making it difficult for other users to establish their purchase decision. A review voting mechanism, in which users can vote for or against a review, addresses this issue (as helpful or not). The helpful votes on a review reflect its usefulness to other users. As voting on usefulness is optional, not all reviews receive this vote. Furthermore, reviews posted recently by users are not associated with any vote (s). The aim of this paper is to predict the usefulness of user reviews through machine learning techniques. Using the Amazon product review dataset of cell phones, classification models are built on eight features and compared on seven performance measures. As per results, all the classification models performed well, except Linear Discriminant Analysis. The classification performance of Logistic Regression, Decision Tree, Random Forest, AdaBoost, and Gradient Boost was unaffected by feature selection or outlier removal. The performance of Linear Discriminant Analysis improved after feature selection but decreased after outlier removal, whereas ET and KNN classifiers improved in both cases. © 2023 Slovene Society Informatika. All rights reserved.