Using Tags and Latent Factors in a Food Recommender System

Due to the extensive growth of food varieties, making better and healthier food choices becomes more and more complex. Most of the current food suggestion applications offer just generic advices that are not tailored to the user's personal taste. To tackle this issue, we propose in this paper a novel food recommender system that provides high quality and personalized recipe suggestions. These recommendations are generated by leveraging a data set of users' preferences expressed in the form of users' ratings and tags, which signal the food's ingredients or features that the users like. Our empirical evaluation shows that the proposed recommendation technique significantly outperforms state-of-the-art algorithms. We have found that using tags in food recommendation algorithms can significantly increase the prediction accuracy, i.e., the match of the predicted preferences with the true user's preferred recipes. Furthermore, our user study shows that our system prototype is of high usability.