Wineinformatics: Wine Score Prediction with Wine Price and Reviews

Wineinformatics is a new field that applies data science to wine-related data. The goal of this paper is to determine whether incorporating wine price can improve the accuracy of score prediction. To explore the relationship between wine price and wine score, naive Bayes classifier and support vector machine (SVM) classifier are employed to predict the scores as either equal to or above 90 or below 90. The price values are normalized using four different methods: mean, median, boxplot mean, and boxplot median. To conduct a proper comparison, the original dataset from previous research, which includes a total of 14,349 wine reviews, was preprocessed by filtering all null price values, resulting in 9721 wine reviews. Using this dataset, classifiers, and normalization methods, the models with and without the price feature were compared. SVM classifier with mean normalization method (USD 50.04) achieved the best accuracy of 87.98%, while naive Bayes classifier with boxplot median normalization method (USD 28.00) showed the greatest improvement of 0.99%. From all the results, we concluded that boxplot median normalization (USD 28.00) is the most effective method in this study. These results indicate that incorporating price as an attribute enhances machine learning algorithms’ ability to recognize the correlation between wine reviews and scores.