Nutrition-guided recipe generation using knowledge graph-enhanced llms

Overview

This cutting-edge AI framework combines the power of food knowledge graphs (KGs) with large language models (LLMs) to create personalized, nutrition-guided recipe recommendations. By integrating structured data from KGs into LLMs, this solution generates recipes that not only satisfy user constraints but also provide detailed micro-nutritional information. This approach significantly outperforms existing methods, offering a cohesive solution for food recommendation, recipe creation, and nutritional analysis.

Technical specifications

• Food Knowledge Graphs: Utilizes FoodKG, a comprehensive database with 67 million triples, to provide structured information about food, nutrition, and regulations.
• LLM Integration: Enhances recipe generation by feeding relevant subgraphs as context into LLMs, ensuring personalized and accurate results.
• Advanced Techniques: Employs multimodal representation learning, Graph-based RAG retrieval, and nutrition prediction fine-tuning to improve the quality of recommendations and dietary compliance.
• Food Substitution Logic: Incorporates algorithms for ingredient substitutions tailored to diverse user queries.

Technology readiness level

Currently at TRL 5, this technology has been validated through component and/or breadboard validation in a relevant environment and is poised for further development and testing.