The Coaching Scenario: Recommender Systems with a Long Term Goal. A Case Study in Changing Dietary Habits

This PhD work explores the way automated recommender systems can be built to help users develop healthier food consumption habits. The main focus is on developing methods of recommendation that allow long-term modifications in user's consumption habits and lasting changes in user's behaviour. We proposed a recommendation scenario were the user and the recommender can be seen as two agents interacting with each other. We also proposed a reinforcement learning formalism of the recommendation problem faced by the recommender system, as well as a choice criterion for recommendation and heuristics derived from that optimal criterion to guide recommendation.