We are happy to share that the paper, “FoodBench-QA: Overview of the Shared Task on Grounded Food and Nutrition Question Answering”, has been presented at the CL4Health Workshop collocated with LREC-COLING 2026 🇪🇸

The work was presented by researchers from the Jožef Stefan Institute, bringing together expertise from the AutoLearn-SI, AutoLLMSelect, AI4Food, and AI4Sci initiatives.

Tome Eftimov, Ana Gjorgjevikj, Matej Martinc, Gjorgjina Cenikj, Sašo Džeroski, and Barbara Koroušić Seljak organized the FoodBench-QA 2026 shared task to advance grounded food and nutrition question answering using food composition databases and semantic food ontologies.

The shared task covered: 🥗 Nutrient estimation from recipes under EU Regulation 1169/2011 tolerance thresholds 🚦 FSA traffic-light classification for fat, sugars, salt, and saturates 🔗 Food named entity recognition and linking to Hansard, FoodOn, and SNOMED CT ontologies

We were very pleased to see participation from five teams across the different tasks and strong results across multiple challenges: 📈 Up to 93.57% accuracy for protein estimation 📈 Macro F1 scores up to 0.90 for traffic-light classification 📈 Named entity linking performance reaching up to 95.75% on artificial datasets

The discussions at the workshop highlighted the growing importance of trustworthy, explainable, and semantically grounded AI systems for food and nutrition intelligence, particularly with the increasing adoption of LLMs and retrieval-augmented systems in healthcare and food domains.

Congratulations to all collaborators and participating teams for contributing to this exciting research direction!