Using Neural Network Models for Wine Review Classification

Abstract Wines are usually evaluated by wine experts and enthusiasts who give numeric ratings as well text reviews. While most classification studies have been based on conventional statistical models using variables, there has very limited work implementing neural network In this paper, we apply (CNN, BiLSTM, BERT) to extract useful information from reviews classify wines according different rating classes. Using a large collection of Wine Spectator , the study shows that BERT, framework recently developed Google, best performance. two-class (90–100 80–89), BERT achieves an accuracy 89.12%, followed BiLSTM (88.69%) CNN (88.02%). four-class (95–100, 90–94, 85–89, 80–84), yields 81.57% accuracy, while other two produce 80% accuracy. The in paper independent domain knowledge thus can be easily extended kinds analysis. Expanding review studies, these up-to-date provide valuable additions data (JEL Classifications: C45, C88, D83)