<![CDATA[Introduction/Main Objectives: Tourism is a vital sector supporting Indonesia’s economic growth, making the effective utilization of public feedback essential for improving service quality. Most feedback is collected through web-based forms in the form of open-text responses that provide rich insights but remain underutilized due to their unstructured nature. Background Problems: This study examines the challenge of identifying the most suitable topic modeling and sentiment analysis techniques for analyzing medium- and long-text feedback in the Indonesian tourism context. Novelty: The novelty lies in the comparative evaluation of classical topic modeling algorithms against modern embedding-based approaches combined with multiple Indonesian transformer models, which has not been extensively explored in tourism-related datasets. Research Methods: The research compares LDA and NMF with BERTopic, Top2Vec, kBERT, and kUSE using coherence scores, and evaluates sentiment analysis using majority voting across transformer architectures. Finding/Results: The results show that BERTopic performed best for medium-length text, while NMF was optimal for long text, and a RoBERTa-based model achieved the highest sentiment agreement. Positive sentiment often appeared in feedback on facilities and fees, whereas negative sentiment dominated topics on environmental and governance issues. These findings offer valuable insights for tourism managers and policymakers in prioritizing improvements and refining strategies.]]>
