COMPARATIVE SENTIMENT ANALYSIS OF VISITOR REVIEWS FOR WATERBOM BALI TOURIST ATTRACTION ON TRIPADVISOR SOCIAL MEDIA USING RANDOM FOREST AND NAÏVE BAYES CLASSIFICATION

With the advancement of technology, especially the internet, the role of the internet as the primary source of information in global life is becoming increasingly crucial. This is particularly true in the context of searching for information about tourist destinations before visiting them. TripAdvisor is a website designed for searching travel destinations and attractions. On this platform, users can provide reviews and see comments from other travelers regarding various tourist destinations, including Waterbom Bali. To gain insights into visitors' perspectives and enhance services for them, the overwhelming number of reviews can be analyzed for sentiment to understand whether travelers' views tend to be positive, negative, or neutral. In this research, the Random Forest and Naïve Bayes methods are employed to conduct sentiment analysis. Scraping data from Waterbom Bali resulted in a dataset of 5750 entries. Despite data imbalance after labeling positive, negative, and neutral sentiments, class imbalance techniques will be applied. The sentiment analysis method, comparing Random Forest and Naïve Bayes, is implemented using the Word2Vec feature extraction method to evaluate its effectiveness. Experimental results show significant differences between the two methods. In Random Forest, after undersampling, an accuracy of 24% was obtained, while oversampling resulted in an accuracy of 98%. Meanwhile, for Multinomial Naïve Bayes, after undersampling, an accuracy of 36% was achieved, and oversampling yielded an accuracy of 97%.