Comparative Study of the Performance of Naïve Bayes, SVM, and K-NN Algorithms for Sentiment Analysis and Topic Modeling of #KaburAjaDulu Hashtags

Abstrak

The #KaburAjaDulu hashtag phenomenon that has been widely discussed on platform X reflects the increasing anxiety of Indonesia's younger generation towards socio-economic conditions and the direction of state policy. This research aims to assess public perception of the hashtag through sentiment analysis and topic modeling approaches. Data was collected from X users' tweets from May to June 2025. The methods used include text preprocessing, sentiment classification using Naïve Bayes, SVM, and K-NN algorithms, and topic modeling with Latent Dirichlet Allocation (LDA). The analysis results show that SVM performs best with 98.93% accuracy and optimal precision-recall balance. The Naïve Bayes model also shows competitive results but tends to favour positive classes. In contrast, K-NN showed the lowest performance due to its inability to overcome the curse of dimensionality in TF-IDF representation. LDA topic modeling identified three main themes: the employment crisis, distrust of institutions due to corruption, and the nationalism vs. migration dilemma. These three topics indicate deep psychological conflicts experienced by youth. The findings support the Self-Determination Theory, which emphasizes the importance of autonomy, competence, and social connection for individual attachment to the environment. Lack of fulfilment of these needs triggers migration intentions as a form of escape or adaptive strategy. This research provides a practical contribution to designing HR policies based on social data. In addition, this approach can be used as the basis for a real-time public perception monitoring system.