A Deep Learning Approach for Tourism Destination Recommendation Using IndoBERT and TF-IDF

Abstract

The rapid development of information technology has transformed various sectors, including tourism, where recommendation systems play a vital role in providing personalized services. Tourists are often faced with a wide range of destination choices, making decision-making increasingly complex. To address this, Artificial Intelligence (AI) and Natural Language Processing (NLP) can be leveraged to enhance recommendation accuracy through deeper analysis of destination descriptions. This study proposes a tourism destination recommendation system combining IndoBERT, SimCSE, and TF-IDF methods. IndoBERT was applied to capture semantic and contextual meaning in the Indonesian language, SimCSE improved sentence-level embeddings, and TF-IDF extracted essential keywords from descriptions. The system was implemented on a website to generate personalized recommendations based on user input. Evaluation results demonstrated that the composition of IndoBERT and TF-IDF achieved strong performance, with precision, recall, and F1-score values of 1.0 at a similarity threshold of 0.20. However, higher thresholds reduced recall and F1-score, indicating that a lower threshold provided a better balance between accuracy and coverage. The recommendation outputs matched user preferences, and functional testing showed that all website features performed successfully. These findings highlight the effectiveness of combining semantic and keyword-based methods for tourism recommendation. Future work could expand the dataset, integrate user feedback, and benchmark against other state-of-the-art models to further enhance system performance.