Tianyi Hu, Andrea Morales-Garzón, Jingyi Zheng, Maria Maistro, Daniel Hershcovich. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Supporting users in their food choices for mindful eating has become one of the spotlights for investigating modern food systems. However, building integral food platforms for this purpose is challenging due to dealing with heterogeneous data sources of different scopes, such as recipe data, food data, and user and dietary specificities. This research paper presents a versatile multi-platform architecture based on micro-services for dealing with different food-related tasks. The contributions of this research are manifold: (1) Firstly, we propose an architecture that enables us to handle various food-related tasks while managing various food data sources, providing data standardisation, scalability, and security benefits; (2) We introduce a novel recipe adaptation algorithm based on intelligent search in external resources and intelligent adaptation of the recipe preparation; (3) We include AI2Cuisine, a mobile application for recipe adaptation to meet different requirements like preferences, health and sustainable goals; (4) Finally, we perform an analysis and discussion in term of the necessity and impact of such sort of applications on the population. To demonstrate the feasibility of our proposal, we have conducted an experimental evaluation, and the results have been validated for various end-users with different expertise. • We present a multi-platform and microservices-based architecture for healthy nutrition food systems. • We introduce a novel and intelligent recipe adaptation algorithm. • We re-train the RoBERTa model to the food-domain. • We present a mobile application (AI2Cuisine) for adapting recipes considering dietary preferences and allergies. • Our study showcases the necessity and potential impact of such applications.

This paper presents AdaptaFood, a system to adapt recipes to specific dietary constraints. This is a common societal issue due to various dietary needs arising from medical conditions, allergies, or nutritional preferences. AdaptaFood provides recipe adaptations from two inputs: a recipe image (a fine-tuned image-captioning model allows us to extract the ingredients) or a recipe object (we extract the ingredients from the recipe features). For the adaptation, we propose to use an attention-based language sentence model based on BERT to learn the semantics of the ingredients and, therefore, discover the hidden relations among them. Specifically, we use them to perform two tasks: (1) align the food items from several sources to expand recipe information; (2) use the semantic features embedded in the representation vector to detect potential food substitutes for the ingredients. The results show that the model successfully learns domain-specific knowledge after re-training it to the food computing domain. Combining this acquired knowledge with the adopted strategy for sentence representation and food replacement enables the generation of high-quality recipe versions and dealing with the heterogeneity of different-origin food data.