Automated Identification of Food Substitutions Using Knowledge Graph Embeddings

Healthy eating is a daily challenge for many, which is in uenced by various factors such as taste, accessibility, price, and the food environment. Consumers often are insufficiently informed about healthier options for the foods they consume. Being able to identify healthy alternatives for foods according to similarities in nutritional value will help consumers choose products that they prefer. This work aims to identify healthy alternatives to foods that also have similar nutritional characteristics through the use of knowledge graph embeddings (KGEs). The quality of the KGEs is assessed against a newly created ground truth, which is verified by two domain experts. Hence, this work presents a newly created ground truth food substitution data set and describes the development of a food recommender system that identifies healthier alternatives to foods.