Seleksi Fitur Information Gain pada Klasifikasi Ulasan Traveloka di Google Play Store Menggunakan Naïve Bayes

Abstract

This study compares sentiment analysis of Traveloka application user reviews using the Multinomial Naïve Bayes algorithm under two scenarios: with Information Gain feature selection and without feature selection. The research data were collected through web scraping from the Google Play Store using the google-play-scraper, resulting in a total of 500 Indonesian-language reviews. Sentiment labeling was conducted using two approaches: manual labeling by a language expert and automatic labeling by the researcher. The collected data then underwent preprocessing stages including data cleaning, tokenizing, case folding, stopword removal, and stemming. In the first scenario, feature selection was applied using the Information Gain method to identify the most influential words for sentiment classification, while in the second scenario a Bag of Words approach without feature selection was employed. Model evaluation was performed using 5-Fold Cross Validation with accuracy, precision, recall, f1-score, and confusion matrix as evaluation metrics, by comparing the model predictions with expert-labeled data. The results show that the Naïve Bayes model with Information Gain feature selection achieved an average accuracy of 87.4%, while the model without feature selection achieved an accuracy of 86.0%. These findings indicate that the application of Information Gain feature selection improves both the performance and efficiency of the sentiment classification model.