How Tasty Is This Dish? Studying User-Recipe Interactions with a Rating Prediction Algorithm and Graph Neural Networks

Andrea Morales-Garzón, Roberto Morcillo-Jiménez, Karel Gutiérrez-Batista, María J. Martín-Bautista

Abstract: Food computing has gained significant attention in recent years due to its direct relation to our health, habits, and cultural traditions. Food-related data have been extensively studied, and graph-based solutions have emerged to combine user-recipe data for various purposes, such as recipe recommendation and food-data alignment tasks. In this study, we propose a graph-based approach to predict the rating a specific user would give a recipe, harnessing the structured form of user-recipe interaction data. The approach incorporates two additional features into the user-recipe interactions graph: 1) user-recipe review embeddings generated by a sentence-based transformer model and 2) a selection of healthy recipe features inferred from nutritional content and international nutrition standards. Results obtained from experiments on a publicly available dataset demonstrate that the proposed method achieves competitive performance compared to recent advancements.