[Background and aims] The KEPHENOL project aims to explore the benefits of polyphenols consumption in mitigating cardiovascular and cognitive decline associated with aging. We have crafted a comprehensive dietary intervention centered on the consumption of 93 key foods, selected from Food Frequency Questionnaires (FFQ) given to the 150 volunteers during the first visit of the study. To accurately assess the participants’ dietary intake, access to various Food Composition Databases (FCDB) is essential. Bringing together heterogeneous required information is challenging, demanding substantial time and effort from researchers.

[Methods] Our proposed methodology integrates different Artificial Intelligence (AI) techniques for creating a machine learning model. The latter employs syntactic analysis of food labels, bolstered by a fuzzy search mechanism leveraging dictionaries. Furthermore, we explored Deep Learning (DL) models to improve the accuracy of the task. These different approaches were integrated to create a comprehensive pipeline for matching food labels among the selected FCDBs.

[Results] Resorting to fuzzy string matching alongside syntactic analysis yielded promising results for raw products. Additionally, leveraging DL sentence embeddings allowed us to recommend related foods by enhancing the original text with semantic similarity. The 93 foods chosen underwent processing to obtain their corresponding codes in FoodEx2, BEDCA, and PhenolExplorer2 databases. Obtained codes were evaluated and manually revised to validate the final codes.

[Conclusion] We present a novel methodology that employs semantic text similarity for automating code selection within FoodEx2, BEDCA and PhenolExplorer2. Results suggest that integrating innovative AI methods into the field of nutritional intervention can streamline researchers’ daily tasks.

Resumen del trabajo presentado al XXXIII Congreso de la Sociedad Española de Nutrición, celebrado en Granada del 19 al 21 de junio de 2024.

Peer reviewed