Traditional recommender systems, such as collaborative filtering, content-based filtering, and hybrid approaches, are limited by challenges including data sparsity and cold start. In order to alleviate these issues, graph-based systems have been increasingly developed for serving recommendations. We build on these existing graph-based approaches and further increase recommendation quality by reflecting the dynamically changing and sequential nature of the recommendation problem and by training prediction models using reinforcement learning (RL). We implement this system using the widely known Netflix Prize data set and build a movie recommender system as a case study. We present results and challenges and discuss how these recommendations can be easily adapted for other user-item interactions as well.