Content-Based Restaurant Recommendation System Using the Best Match 25 Lucene (Bm25l) Method

Abstract

Choosing a suitable dining place can be challenging for both locals and tourists due to the wide variety of culinary options and the limited availability of comprehensive recommendations covering food types, menu variations, prices, facilities, operating hours, and locations. To address this issue, this study aims to develop a Content-Based Restaurant Recommendation System Using the Best Match 25 Lucene (BM25L) Method in Pamekasan that assists users in selecting dining places aligned with their preferences. The system leverages the Best Match 25 Lucene (BM25L) method, which effectively accounts for document length in ranking, providing more precise recommendations. Data were collected from public sources such as Google and Instagram, complemented by direct field observations. System performance was evaluated through precision@k testing with 10 users, yielding a precision@10 score of 0.93 and a precision@20 score of 0.84. The results indicate that the most relevant recommendations typically appear within the top 10 positions, while relevance slightly decreases as the number of evaluated results increases. This demonstrates that the system is highly accurate for smaller result sets while maintaining acceptable relevance for larger sets. The research contributes to improving user experience by enabling faster and more reliable decision-making when choosing dining venues. Furthermore, the system provides a practical framework for future applications of content-based recommendation methods in the culinary domain and other service sectors. By prioritizing the most relevant options, this system enhances convenience, supports informed choices, and can serve as a model for similar smart recommendation solutions in regional tourism and hospitality contexts.