<![CDATA[The development of Indonesia’s New Capital City (IKN) has generated diverse public responses on social media, particularly YouTube, making sentiment analysis necessary to map public perceptions. Previous studies have reported relatively low classification accuracy, reaching only 60%, indicating the need for more effective approaches to improve performance. This study aims to compare the performance of the Naïve Bayes algorithm in classifying public sentiment toward the IKN development using two feature extraction methods, namely TF-IDF and Word2Vec. The data were collected from YouTube comments and processed through preprocessing stages, expert-based labeling, and evaluation using 10-Fold Cross Validation. The results show that the TF-IDF-based Multinomial Naïve Bayes model achieves the best performance with an accuracy of 83%, a positive recall of 82%, and a negative F1-score of 85%, outperforming the Word2Vec-based Gaussian Naïve Bayes model, which attains an accuracy of 82% with a lower positive recall of 76%. These findings confirm that TF-IDF is more effective and stable in handling short-text comment characteristics than Word2Vec, which requires a larger corpus for optimal semantic representation.]]>
Copyrights © 2026
