COMPARISON OF SENTIMENT ANALYSIS USING NAÏVE BAYES CLASSIFICATION METHOD AND LEXICON BASED ON JIWA+ BY JANJI JIWA APPLICATION REVIEWS

Abstract

The coffee beverage industry in Indonesia is experiencing significant growth, intensifying competition among businesses striving to maintain quality for customer loyalty. E-commerce applications play a vital role in preserving business standards as they directly engage with consumers. Janji Jiwa is among the coffee brands leveraging an application named Jiwa+ in their operations. Analyzing reviews on this e-commerce platform provides valuable insights for business owners and app developers. In this study, sentiment analysis was conducted by classifying reviews into positive, neutral, and negative sentiments using two methods: Lexicon Based and Naïve Bayes. The Lexicon Based method uses a predefined dictionary as the basis for labeling, while Naïve Bayes relies on training data to provide new insights into how both methods handle this type of data. A total of 597 Jiwa+ application reviews from the Google Play Store were utilized, split into 90% training and 10% testing data sets. The study results indicate that Naïve Bayes produces a better model than the Lexicon-Based method, as shown by its higher accuracy, sensitivity, and specificity. This is because Lexicon-Based relies on labeling words from a dictionary, which may not cover all words in the reviews, leading to labeling errors and misclassification.