The purpose of our research is to study reinforcement learning approaches to building a movie recommender system. We formulate the problem of interactive recommendation as a contextual multi-armed bandit, learning user preferences recommending new movies and receiving their ratings. We show that using reinforcement learning solves the problem of exploitation-exploration trade-off and the cold-start problem. We integrate the novelty of movies to the model. We explore a content based approach as well as a collaborative filtering approach and both yield viable recommendation results.