Development of a Context-Dependent Recipe Recommendation System Using Graph Convolutional Networks and Contrastive Learning

Abstract

This study proposes a method combining graph convolutional networks (GCN) and contrastive learning to construct a context-dependent recipe recommendation system. The proposed method focuses on recipes and integrates nutritional content, ingredient co-occurrence relationships, and visual features of dishes to achieve recommendations tailored to the user's context. In the evaluation experiment, we compared the proposed method with conventional popularity-based recommendations, demonstrating the superiority of the proposed method in specific contexts. This research aims to contribute to supporting users' healthy dietary habits through context-dependent recipe recommendations.