Graph‐based food recommendations

Abstract

Recipe recommendation systems play an important role in assisting users to find food recipes that meet their requirements. Pioneering research on the use of Graph Neural Networks (GNNs) for food recipe recommendation systems has produced superior results when compared to contemporary recommendation system performance benchmarks. This research is focused at the intersection of three traditionally separate fields of study namely, Recommendation Systems, Graph Representation Learning using GNNs and Food Recipe Heterogeneous Graphs. The aim is to investigate the suitability of GNNs for recipe recommendations. The first objective is to construct a heterogeneous graph representing the interconnections between recipes, ingredients, and users. This graph is constructed from a publicly avail‐ able dataset and stored in a graph database. The graph properties are then analysed using graph techniques. The second objective is to explore the use of GNN propagation models that have been designed for heterogeneous applications, on the performance of a food recipe recommendation system through a set of experiments. The performance of the GNN recommendation system when a health category attribute is added to recipe nodes is also explored. The high Test AUC‐ROC results obtained demonstrated the potential of GNN models to be used for food recipe recommendation systems using heterogeneous graph information. It was also demonstrated that the addition of health category information attributes to the recipe nodes to enhance the ingredient attribute information did not improve the results obtained.