The heterogeneous tourism landscape of Labuan Bajo, ranging from marine excursions to cultural sites, is often inadequately captured by conventional retrieval systems due to vocabulary mismatch. The reliance on rigid lexical matching leads to a failure in grasping the search intent of tourists who use terminology that differs from, yet is semantically equivalent to, destination descriptions. This study addresses this limitation by developing a hybrid recommendation model enhanced with a Semantic Feature Boosting approach. The proposed methodology integrates lexical feature extraction based on TF-IDF (1,2 n-grams) and deep semantic representation using the Sentence-BERT model (paraphrase-multilingual-MiniLM-L12-v2). Both vector representations are combined through an adaptive weighting scheme (0.6 for TF-IDF and 0.4 for Sentence-BERT) to compute destination relevance based on cosine similarity. Empirical evaluation across ten query scenarios representing diverse tourist needs demonstrates robust performance: Mean Precision@5 = 0.90, Mean Recall@5 = 0.81, Mean Average Precision@5 = 0.89, and Mean NDCG@5 = 0.93. Statistical significance testing (Wilcoxon signed-rank, p < 0.05) confirms that the hybrid approach outperforms Pure lexical and Pure semantic baselines. This comprehensive evaluation demonstrates that the integration of semantic features significantly enhances the system's ranking quality and ability to understand diverse search contexts. Thus, the proposed approach proves effective in generating more precise and adaptive recommendations for complex tourism landscapes, while providing a methodological contribution to recommendation systems responsive to natural language variations.
Copyrights © 2026