Researchers Develop AI Method for More Personalized Nutrition Guidance
Researchers from the AIRI Institute and ITMO University have introduced PA-CoT, a method that improves the quality of personalized dietary recommendations generated by AI models. The approach reduces generic or potentially risky advice by enabling a more thorough analysis of individual user data.
General dietary recommendations may be irrelevant—or even inappropriate—for people with different physiological characteristics. Advice that is suitable for a healthy individual may need to be adjusted for people with chronic conditions, allergies, or specific health and fitness goals.
PA-CoT operates in several stages. First, the model analyzes the user's profile, identifying health goals, activity level, and dietary restrictions. It then generates recommendations tailored to those factors. In the final step, the system evaluates its response for potential risks and, when appropriate, adds a disclaimer advising the user to consult a physician.
The researchers evaluated the method on 200 user cases with reference answers prepared by dietitians, comparing its performance with 11 alternative approaches. On a five-point scale, PA-CoT scored above 4.7 for personalization and above 4.6 for safety, outperforming the closest competing method. The team plans to expand the evaluation dataset and develop a conversational AI assistant for nutrition guidance.
