Food Pairing Based on Generative Adversarial Networks

Abstract

The Generative Adversarial Networks (GAN) has received great attention and achieved great successes in many applications. It is still being intensively developed and get many different variants of GAN. GAN was proposed to generate similar-looking samples to those in the training data sets. The emergence of GAN and its variants also provide new ideas for food pairing. In this paper, we have tried to invent a novel technique for food pairing using GAN and its variants. Specifically, we adopted the Long Short Term-Memory (LSTM) as the generator and the Convolutional Neural Network (CNN) as the discriminator. The sequences of recipes as the input will be encoded by LSTM into target sequences, which were finally identified by CNN to compute the differences between the generated recipes and their original input. The CNN will give a feedback to LSTM to optimize its parameters until the end of training process. As different customers have different food tastes, we have improved our method and invented new model using Conditional GAN (CGAN) to incorporate the personal demands in food pairing. We have conducted extensive experiments on real data sets to evaluate the efficiency of our proposed methods. The experimental results proved that our methods can generate better food pairings.