z-Batista, K., & Martin-Bautista, M. J. (2024). Link prediction in food heterogeneous graphs for personalised recipe recommendation based on user interactions and dietary restrictions Morales Garzón, Andrea Martín Bautista, María José Gutiérrez Batista, Karel Link prediction Heterogeneous graph Food computing Graph neural network Recipe data and user interactions and preferences have been widely studied in food computing, especially for the recipe recommendation task. One part of these works seeks to introduce healthy patterns while considering user preferences, known as healthy-aware recommender systems. The major challenge here is to build systems capable of learning the complex structure of recipe data since they involve heterogeneous resources. Internet-sourced recipe collections may also have a representative amount of recipes that do not follow healthy guidelines, thus inhibiting the performance of health-aware recommender systems. We propose a new method for recipe recommendation based on a link prediction algorithm that considers recipes, their healthy features, and users. We train the model twice, once with the whole dataset and once with recipes following healthy guidelines. We follow three strategies for representing recipe data regarding healthy features. In general, training the model in recipe data that follows healthy guidelines achieves better results, especially when representing recipes with numeric nutrition recipe values. 2025-06-13T08:58:08Z 2025-06-13T08:58:08Z 2023-11-15 journal article Morales-Garzón, A., Gutiérrez-Batista, K., & Martin-Bautista, M. J. (2024). Link prediction in food heterogeneous graphs for personalised recipe recommendation based on user interactions and dietary restrictions. Computing, 106(7), 2133-2155. https://hdl.handle.net/10481/104628 eng http://creativecommons.org/licenses/by-nc-nd/4.0/ open access Attribution-NonCommercial-NoDerivatives 4.0 Internacional Springer