Can reviews predict reviewers’ numerical ratings? The underlying mechanisms of customers’ decisions to rate products using Latent Dirichlet Allocation (LDA)

Abstract

Purpose Two valuable pieces of information – reviews and their corresponding numerical ratings – are accessible to potential customers before they make a purchasing decision. An extensive body of marketing literature has scrutinized the influence of customers’ reviews by linking such aspects as the volume and valance of reviews with product sales and customers’ purchase intention. The aim of this study, for which dual coding theory was used, was to understand the relationship between reviews and their corresponding numerical ratings. Design/methodology/approach The authors used the latent Dirichlet allocation technique to categorize customers’ reviews. The present findings contribute to the literature by showing the underlying mechanisms that customers use to interpret reviews and associate them with numerical ratings. Findings The gradient boosted decision tree model demonstrates that non-abstract-dominant reviews (reviews mainly consist of tangible objects, actions, events or affective words) are significant predictors of their corresponding numerical ratings. However, abstract-dominant reviews (i.e. those consisting primarily of intangible objects, events or actions) cannot predict their associated numerical ratings. Originality/value The present findings contribute to the literature by showing the underlying mechanisms that customers use to interpret reviews and associate them with numerical ratings.