COMPARATIVE EVALUATION OF SVM AND LSTM FOR TOURISM SENTIMENT CLASSIFICATION: STUDY CASE TANAH LOT TOURISM DESTINATION, BALI

Abstract

This study presents a comparative evaluation of Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) models for tourism sentiment classification, using YouTube comments related to Tanah Lot, Bali. The dataset manually cleaned comments labeled as Positive, Neutral, or Negative. Both models achieved identical overall accuracy (0.95), but class-wise analysis revealed substantial differences: LSTM exhibited strong bias toward the majority class (Neutral), failing to detect minority sentiments, while SVM retained partial sensitivity to Positive and Negative classes. These findings highlight the limitations of deep learning architectures under low-resource and imbalanced conditions and underscore the importance of context-aware model selection. Class-wise evaluation metrics are essential for capturing minority sentiment, which is critical for destination governance and informed decision-making in tourism management.