Understanding traveler feedback is essential for enhancing tourism services and improving visitor experiences. This paper proposes a framework that leverages large language models (LLMs), specifically GPT-4o mini, to perform sentiment classification on multilingual travel reviews. As the primary innovation, GPT-4o mini generates accurate sentiment labels for each review, capturing users’ emotional responses toward various aspects of their travel experiences. Complementing this, the framework uses semantic topic matching by embedding reviews with the all-miniLM-L6-v2 Sentence Transformer and comparing them to predefined tourism-related categories using FAISS for scalable similarity search. The proposed framework is applied to a multilingual dataset of reviews from Google Maps and TripAdvisor, focusing on Kyoto, Japan. Results demonstrate that integrating LLM-based sentiment classification with embedding-driven topic matching enables precise, interpretable insights, empowering tourism stakeholders to identify key areas of concern and enhance destination management strategies.