Dietary Behavior Based Food Recommender System Using Deep Learning and Clustering Techniques

Abstract

Deep learning algorithms have been highly successful in various domains, including the development of collaborative filtering recommender systems. However, one of the challenges associated with deep learning-based collaborative filtering methods is that they require the involvement of all users to construct the latent representation of the input data, which is then utilized to predict the missing ratings of each user. This can be problematic as some users may have different preferences or interests, which may affect the accuracy of the prediction generation process. The research proposed a food recommender system, which tries to find users with similar dietary behavior and involve them in the recommendations generation process by combining clustering technique with denoising autoencoder to generate a rate prediction model. It is applied to “Food.com Recipes and Interactions” dataset. RMSE score was used to evaluate the performance of the proposed model which is 0.1927. It outperformed the other models that used autoencoder and denoising autoencoder without clustering where the RMSE values are 0. 4358 and 0.4354 consequently.