Analysis of Cosmetic Product Opinions on E-commerce Based on Naïve Bayes Classifier

Abstract

This research employs Naïve Bayes classifier to analyze consumer sentiment in concealer reviews, utilizing a dataset of 400 reviews as training data, and testing it to 100 reviews of five different brands on their concealer product. The model’s performance achieved an accuracy of 89.37% on training data and 73.75% on the testing data. A dictionary was created highlighting twelve frequently used words in the reviews, along with their frequencies and positive probabilities across five brands, offering insights for improvement. Additionally, the study presents a confusion matrix detailing precision and recall for each brand. The results indicate that Brand D performed the best, followed by Brands B, C, E, and A, providing recommendations for enhancing customer satisfaction based on sentiment analysis.