A Deep Learning Model For Food Pairing

Abstract

Humans have come a long way, from nomads to farmers, from growing fruits and livestock farming for their survival to having the luxury of innovating with food that produced different types of cuisines and became the identity of different cultures. Food science, multidisciplinary science that studies food’s physical, biological and chemical aspects, has been around for centuries. One emerging aspect is the study of food pairing. Chefs have tested countless food ingredient pairs through trial and error and using their expertise in their respective cuisine styles, but this method is finite and consumes energy and resources. In this study, we proposed two approaches based on deep learning techniques to create a model that predicts the scores of ingredient pairs. The first approach employs a Siamese Neural Network model that recommends ingredient pairs using the frequency of appearance of those pairs. The second approach focuses on recommending ingredient pairs that share similar flavor compounds. We have concluded that both models give us insights on how to innovate regarding pairing food ingredients. Where the first one considers familiar ingredient pairs, the second one recommends uncommon new pairs based on the food pairing hypothesis, which states that food with similar flavor compounds tastes good when consumed together. Keywords: deep learning, food pairing, siamese neural network, food pairing hypothesis, natural language processing