Most recommender systems recommended a list of K items, such as restaurants, songs, or movies. The user examines the recommended list from the first item to the last and typically clicks on the first item that attracts the user and does not examine rest of the items. The cascade model is a simple, intuitive & popular model to formulate this kind of user behavior. In this project, we implement and empirically evaluate the performance of an online learning variant of the cascade model, which is known as linear cascading bandits proposed by Zong Shi, et al. on movie recommendation problem. We will implement two algorithms CascadeLinTS and CascadeLinUCB for solving this problem which are based on the idea of linear generalization i.e. we assume that attraction probabilities of items are a linear function of the known features of items $x_e$ and an unknown parameter vector θ and as a result both algorithms gives regret which does not depend on the number of items L.