Labuan Bajo Culinary Tourism Recommendation System: A Comparison of Content-Based Filtering Similarity Methods

Abstract

Labuan Bajo, as a national super-priority tourism destination, has experienced a significant increase in tourist visits, dominated by foreign tourists, which has driven the expansion of the culinary sector with a continuously increasing number of restaurants. This makes it difficult for tourists to choose dining options that match their preferences among the many available choices. Previous culinary tourism recommendation system research has only used a single similarity measurement technique without evaluating its effectiveness compared to other techniques. This study aims to build a culinary tourism recommendation system in Labuan Bajo using the Content-Based Filtering method, as well as to compare three similarity measurement techniques Cosine Similarity, Euclidean Distance, and Jaccard Similarity to determine which technique produces the most relevant recommendations. Data was obtained through scraping techniques on 150 restaurants in Labuan Bajo, covering category and rating attributes, which were then processed into a feature matrix for inter-item similarity calculation. Evaluation was carried out using Precision@5 with relevance criteria based on category similarity and rating proximity, followed by a Wilcoxon Signed-Rank Test to examine statistical significance between methods. The test results show that Cosine Similarity and Euclidean Distance produce an equal precision of 0.6947, higher than Jaccard Similarity at 0.6320, with the difference proven statistically significant (p < 0.001). The system was then implemented as a web application using the CodeIgniter framework, allowing users to select a similarity method and view restaurant recommendations interactively. This study demonstrates that considering the rating attribute, not just category, produces more relevant recommendations than a category-only approach.