<![CDATA[Tourism recommendation systems frequently encounter extreme data sparsity, cold-start problems, and limited personalization when relying solely on either collaborative or content-based approaches, particularly in real-world review platforms where user–item interactions are unevenly distributed. Such conditions reduce prediction reliability and weaken the practical usefulness of recommendation outputs. This study aims to design and empirically evaluate a hybrid conversational tourism recommender that improves recommendation accuracy and ranking relevance under sparse data conditions. The proposed framework integrates collaborative filtering and content-based filtering within a hybrid architecture, combining interaction-driven preference learning with textual content modeling derived from user reviews. The recommendation engine is operationalized through a conversational chatbot interface to support interactive preference elicitation and real-time recommendation delivery. The system is evaluated using a Google Maps review dataset characterized by high sparsity. Model performance is assessed using rating prediction metrics and top-K ranking metrics to ensure comprehensive evaluation. Experimental results indicate that the hybrid model achieves a Root Mean Squared Error of 0.6985 and a Mean Absolute Error of 0.5448, outperforming standalone collaborative and content-based models. In top-10 recommendation evaluation, the model attains Precision@10 of 0.52, Recall@10 of 0.91, and F1-score@10 of 0.66, demonstrating improved relevance and coverage. These findings highlight the effectiveness of hybrid conversational integration in mitigating sparsity-related limitations and provide practical implications for scalable, data-driven personalization in digital tourism platforms.]]>
