PREFer: A prescription-based food recommender system

Abstract

In the literature, several researches on food recommendation and automatic menu generation have been proposed, taking into account different aspects, such as personal and cultural preferences, health and religion constraints, menu composition and recipe co-occurrence. However, recommending recipes and menus, which not only meet the user's preferences, but are also compliant with best food habits, is still an open issue. This paper presents the PREFer food recommender system, apt to provide users with personalized and healthy menus, taking into account both user's short/long-term preferences and medical prescriptions. Prescriptions classify the ideal user's nutrition behaviour from the health point of view, with constraints imposed by the specific user's phenotype. In fact, major novel contribution of the proposed system is the use of prescription types associated with users' profiles to improve users' behaviour in selecting food. The recommended menus are generated through three steps. First, according to user's request, recipes are selected by content-based filtering, based on comparisons among features used to annotate both users' profiles and recipes. Second, candidate menus are generated using the selected recipes. Third, menus are refined and ranked taking into account also prescriptions. The PREFer system has been developed within a regional project, related to the main topics “Feeding the Planet, Energy for Life” of the 2015 World Exposition (EXPO2015, Milan, Italy), where the University of Brescia aimed at promoting healthy behavioural habits in nutrition.