This paper presents a new training method that allows the recommender to learn the user's next intention and more intentions to come.
