Optimizing Sentiment Analysis of Digital Wayang Viewer Comments using SMOTE and the Naïve Bayes Algorithm

Abstract

Wayang performances are an integral part of Indonesia’s rich cultural heritage. This traditional art form has been deeply rooted in Indonesian society for centuries, evolving through live performances and, more recently, through rapid digital adaptations—including presentations on online platforms such as YouTube. In the digital age, YouTube has become a leading platform for video sharing, allowing audiences to enjoy wayang performances without being physically present. However, data from the Central Bureau of Statistics on Socio-Cultural Affairs indicates a decline in interest among younger generations in traditional arts such as wayang. This highlights the need for innovative and relevant approaches to reintroduce this cultural heritage to them. Sentiment analysis based on viewer comments offers an effective way to identify audience opinions—whether positive, negative, or neutral. Comment data were collected using web scraping techniques with Selenium WebDriver, enabling efficient data extraction. The collected data then underwent preprocessing, including case folding, tokenization, and stopword removal, to prepare it for classification. The Naïve Bayes algorithm was employed to categorize comments into positive, negative, or neutral sentiments. Preliminary results revealed that 51.6% of comments were positive, 42.3% neutral, and 6.0% negative. Model evaluation using K-fold cross-validation yielded an accuracy of 0.98 ± 0.01, a precision of 0.99 ± 0.01, and a recall of 0.72 ± 0.11 without applying SMOTE. After applying SMOTE, recall improved to 0.80 ± 0.05. This study contributes to the development of more accurate sentiment analysis models in the context of social media and underscores the importance of techniques like SMOTE in addressing class imbalance issues.