Language models in nutrition and dietetics: a scoping review.

Area of Science:

• Artificial Intelligence in Nutrition Science
• Dietetics and Computational Health

Background:

• Generative artificial intelligence (AI) offers potential for personalized nutrition by understanding and generating text.
• However, AI in nutrition poses public health risks including inaccuracy, bias, and data privacy concerns.
• The extent of research on AI applications in nutrition and dietetics remains unclear.

Purpose of the Study:

• To systematically review the applications of language models in nutrition and dietetics.
• To identify the opportunities and challenges associated with these AI applications.

Main Methods:

• A comprehensive literature search was conducted across 10 major databases (Medline, EMBASE, PsycINFO, CINAHL, PubMed, Global Health, Web of Science, Scopus, DBLP, and IEEE Xplore).
• The review followed a predefined protocol, including forward and backward citation chasing for all included articles.
• Original studies published in peer-reviewed journals were included in the analysis.

Main Results:

• Out of 2118 identified articles, 40 were included, all published between 2022 and 2024.
• Twenty-three different language models were utilized, with 9 studies developing domain-specific models via retrieval augmented generation or fine-tuning.
• Primary applications included dietary advice (21 studies), food composition estimation (9 studies), meal plan creation (7 studies), and data extraction/summarization (5 studies).

Conclusions:

• The application of language models in nutrition and dietetics is an emerging field with limited real-world implementation data.
• Further research is needed in practical settings to build a stronger evidence base.
• Future studies should incorporate perspectives from nutrition professionals, patients, and the general public.