Wine Review Descriptors as Quality Predictors: Evidence from Language Processing Techniques

Abstract There is an ongoing debate on whether wine reviews provide meaningful information properties and quality. However, few studies have been conducted aiming directly at comparing the utility of numeric measurements in data analysis. Based from close to 300,000 wines reviewed by Wine Spectator , we use logistic regression models investigate are useful predicting a wine's quality classification. We group our sample into one two binary brackets, with critical rating 90 or above other ratings 89 below. This outcome constitutes dependent variable. The explanatory variables include different combinations numerical covariates such as price age representations text reviews. By accuracy models, results suggest that review descriptors more accurate classifications than various covariates—including price. In study, three feature extraction methods analysis: latent Dirichlet allocation, term frequency-inverse document frequency, Doc2Vec embedding. find best performing method produces highest classification due its capability using contextual documents. (JEL Classifications: C45, C88, D83)