2026 Volume 77 Issue 3 Pages 151-162
In recent years, the widespread use of the Internet has led many users to search for recipes online. In particular, user-generated recipe websites, where users can freely post their own recipes, have gained popularity, creating an environment in which a wide variety of recipes is readily accessible. However, many of the recipes available on these sites have been criticized for not necessarily being healthy from a nutritional standpoint. Against this backdrop, websites that feature recipes created by professionals, such as registered dietitians, with a strong focus on health, have begun to attract attention. These sites offer the advantage of ensuring both nutritional balance and taste, as the recipes are developed by experts. However, creating recipes with professional supervision incurs substantial costs, significantly limiting the number of available recipes compared to user-generated recipe websites. The limited number of recipes restricts users' options and makes it difficult to meet diverse needs. To address this issue, this study proposes a method for identifying substitute ingredients to efficiently expand the number of recipes on health-conscious recipe websites while maintaining both taste and nutritional balance. Specifically, we assume that expert-created recipes are optimized for both taste and nutrition and aim to select substitute ingredients that do not significantly deviate from the nutritional values of existing recipes. Furthermore, because the target recipe websites contain a relatively small number of recipes, it is challenging to directly apply conventional large-scale data-driven methods. Therefore, this study develops a method that combines substitute ingredient extraction using a large language model (LLM) and human feedback. To evaluate the effectiveness of the proposed method, we conduct an appropriateness assessment by registered dietitians.