Improvement of User Sentiment Classification Model for the Indomaret Poinku Application Using the Naïve Bayes Method

Abstract

The Indomaret Poinku application provides a platform for users to give reviews related to products and services. With the increasing number of user reviews, an effective method is needed to automatically analyze opinions. This research aims to improve the sentiment analysis model on the Indomaret Poinku application using the Naive Bayes algorithm. The selection of this algorithm is based on its simplicity and effectiveness in text classification. To improve the model's performance, this research also applies preprocessing techniques such as cleaning, case folding, tokenizing, normalization, stopword removal, and stemming, as well as the feature selection technique Information Gain. The research method involves stages of collecting review data from the Google Play Store, manually labeling the data, and analyzing the data using TF-IDF numerical representation. The Multinomial Naive Bayes model was trained and tested using evaluation metrics such as accuracy, precision, recall, and F1-score. The evaluation results show that the developed model is capable of achieving an accuracy of 75.5%, a precision of 78%, a recall of 75%, with an average F1-score of 70.4%. Further analysis shows that features such as "good" and "great" have a significant influence in sentiment classification. The results of this study reveal that the enhancement of the Naive Bayes model through feature selection and optimization of the Preprocessing process is capable of improving sentiment classification accuracy. These findings contribute to application developers in understanding user opinions, which can be used to improve the quality of services and products.