Personalised healthy food text recommendations through fuzzy linguistic variables: A generative AI-based approach
Metadatos
Mostrar el registro completo del ítemAutor
Gutiérrez Batista, Karel; Martín Bautista, María José; Morcillo Jiménez, Roberto; Morales Garzón, Andrea; Rojas Carvajal, Ana MaríaEditorial
Elsevier
Materia
Food recommendation Generative models Prompt engineering Natural language processing Fuzzy logic
Fecha
2025-05-12Referencia bibliográfica
Morales-Garzón, A., Carvajal, A. M. R., Morcillo-Jimenez, R., Martin-Bautista, M. J., & Gutiérrez-Batista, K. (2025). Personalised healthy food text recommendations through fuzzy linguistic variables: A generative AI-based approach. Applied Soft Computing, 113234. https://doi.org/10.1016/j.asoc.2025.113234
Patrocinador
Departamento Ciencias de la Computación e Inteligencia Artificial Grupo de investigación IDBIS; MCIN/AEI/10.13039/501100011033 PID2021-123960OB-I00; ERDF A way of making Europe; Junta de Andalucía PREDOC_00298; University of Granada CITIC-2024-06; Universidad de Granada/CBUAResumen
Nutrition and healthy eating habits are fundamental for the global population. Nowadays, there is an increasing tendency to consume less healthy recipes and a low general knowledge of nutrition. In these terms, generative AI arises as a potential tool for health-aware food recommendations, especially when improving communication with the user. This study presents a pipeline to enrich prompts with fuzzy modelling to increase the quality of textual recommendations. We apply our pipeline to generate a personalised frequency of food consumption, considering both nutritional and individual profiles. This is an essential task for increasing the health-conscious recommendation systems. We conducted extensive experimentation across different roles and prompt strategies. We evaluated the quality of the text and the nutritional rigour of the text responses. Our results show that enriching prompts with fuzzy modelling of the nutritional information of the foods significantly improves the quality of the prompt responses.