Abstract
ABSTRACT This study presents a novel approach to enhancing menu recommendation systems by integrating auto-tagging techniques for more personalized and context-aware dining experiences. Using a dataset of over 30,000 restaurant reservations from a South Korean platform (2010–May 2023), we analyzed 15,253 purchase records from 858 users. Users were clustered based on contextual factors like region, reservation time, and order commitment level, revealing 13 distinct dining preference patterns. Our multi-step recommendation process incorporates cluster weights, tag similarity, and past reservation data to generate personalized menu recommendations, particularly for travelers. Compared to a baseline model without tag-based filtering, our approach showed significant improvements in recall and mean average precision (mAP), achieving over a tenfold increase in both metrics. The base model's reliance on menu names limited its ability to analyze similarities accurately, underscoring its limitations. In contrast, our model leverages culinary tags to identify preferences at a granular level, enabling more precise and relevant recommendations. These findings highlight the potential of tag-based clustering to transform dining recommendation systems.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call