Exploring Naive Bayes Variants for Classifying Emotions in e-Commerce Reviews

This research aims to classify emotions from customer reviews on the Tokopedia e-commerce platform using the Naïve Bayes algorithm. The dataset used is a public dataset consisting of 5,400 product reviews across 29 categories, with emotional annotation carried out by a clinical psychologist. Three Naïve Bayes variants were used: NB_Gaussian, NB_Multinomial, and NB_Bernoulli, each tested with and without resampling to address data imbalance. Evaluation was conducted using accuracy, precision, recall, and f1-score metrics. The results show that the NB_Multinomial model with resampling achieved the highest accuracy of 63.33% and the highest precision of 68.13%, while the highest f1-score of 58.46% was achieved by NB_ Bernoulli with resampling. These findings indicate that resampling significantly impacts model performance in emotion classification.