Analisis Sentimen Review Toko Menggunakan Naive Bayes dengan Penanganan Negasi dan Mutual Information

Abstract

Sentiment analysis refers to a technique within natural language processing for extracting and interpreting sentiments or opinions regarding a particular topic, such as product reviews. The model in this study was built through a series of stages, including preprocessing, feature extraction using TF-IDF, feature selection using Mutual Information, and sentiment classification with Multinomial Naive Bayes algorithm. The preprocessing stage consists of several processes, namely cleaning, case folding, tokenization, normalization, stopwords removal, stemming, and handling of negations and intensifiers. This study was tested using four scenarios. In the first scenario, classification was performed using the Multinomial Naive Bayes algorithm as a baseline approach. The second scenario involved classification with the handling of negations and intensifiers during the preprocessing stage. The third scenario involved classification with feature selection using Mutual Information. The fourth scenario combined the handling of negations and intensifiers with feature selection using Mutual Information. Based on the results, the classification using Multinomial Naive Bayes with the combination of negation and intensifier handling and feature selection using Mutual Information delivered the highest performance, attaining 74.08% accuracy, 74.32% precision, 74.19% recall, and an F1-Score of 74.04% with 60% of the features selected.