Proposal for a Novel Recipe Generation Approach Using Inverse TF-IDF and Generative AI

Abstract

This paper proposes a novel recipe generation system that integrates an inverse term frequency–inverse document frequency (TF-IDF) approach with generative artificial intelligence (AI) to balance creativity with practicality in home cooking. While conventional recipe recommendation systems tend to focus on commonplace recipes or staple ingredients, this study adopted an inverse approach by focusing on low-IDF ingredients and leveraging them as the foundation for generating highly reproducible recipes. Utilizing GPT-4, the system automatically constructs prompts that reflect user preferences and cooking conditions, enabling the generation of context-sensitive recipes. A lightweight web-based demo system was developed, allowing users to input two ingredients and receive original recipe suggestions along with cosine similarity scores in real time. Even when users input seemingly unrelated ingredients, the system provides composite recipe outputs, demonstrating its creative flexibility. This approach uncovers the latent creative potential in common ingredients and suggests broad applicability in the food industry, nutritional planning, and culinary education. By combining generative AI with statistical ingredient analysis, the proposed method offers a new intelligent support model for everyday cooking practices.