Apple doesn’t fall far from the tree: Effect of extrinsic factors of online reviews on predicting useless reviews

The surge in useless online reviews has disrupted the ecosystem of review platforms, leading to increased information overload and wasting resources needed to manage such excessive data. Despite extensive research on predicting helpful reviews, little attention has been given to identifying useless reviews, which can be crucial for maintaining and improving the platforms. Drawing from the theory of literature, this study develops predictive models to classify potentially useless online reviews—defined as those that have received no helpfulness votes for over a year—by considering factors from three dimensions: review, reviewer, and product. Utilizing nearly 62 million online reviews from Amazon.com, this study tests machine learning algorithms, including random forest, neural network, gradient boosting machine, and logistic lasso regression, to offer an effective model for detecting useless reviews. Additionally, incorporating reviewer and product factors, less emphasized in existing research, significantly enhances prediction performance. Specifically, reviewer expertise, reviewer experience, and product popularity greatly improve the prediction models. For practitioners, this study underscores the importance of reviewer factors in managing online reviews and provides practical methods to identify potentially useless reviews early on.