Abstract
Food delivery platforms face the challenge of helping users navigate vastcatalogs of restaurants and dishes to find meals they truly enjoy. This paperpresents RED, an automated recommendation system designed for iFood, LatinAmerica's largest on-demand food delivery platform, to personalize theselection of curated food collections displayed to millions of users. Ourapproach employs a LightGBM classifier that scores collections based on threefeature groups: collection characteristics, user-collection similarity, andcontextual information. To address the cold-start problem of recommending newlycreated collections, we develop content-based representations using itemembeddings and implement monotonicity constraints to improve generalization. Wetackle data scarcity by bootstrapping from category carousel interactions andaddress visibility bias through unbiased sampling of impressions and purchasesin production. The system demonstrates significant real-world impact throughextensive A/B testing with 5-10% of iFood's user base. Online results of ourA/B tests add up to 97% improvement in Card Conversion Rate and 1.4% increasein overall App Conversion Rate compared to popularity-based baselines. Notably,our offline accuracy metrics strongly correlate with online performance,enabling reliable impact prediction before deployment. To our knowledge, thisis the first work to detail large-scale recommendation of curated foodcollections in a dynamic commercial environment.