CapDRL: A Deep Capsule Reinforcement Learning for Movie Recommendation

Recommender systems provide users with a personalized list based on individual interests. There are three main challenges in traditional movie recommendation models: (1) considering recommendation procedure as a static one; (2) not taking user’s feedback into consideration; (3) it’s hard to extract similar features of items rated by users effectively. To address these, we propose a Deep Reinforcement Learning method based on the Capsule Network for the movie recommendation, called CapDRL. Roughly speaking, to solve the first two problems, we formulate the task of sequential interactions between users and recommender systems as a Markov Decision Process and automatically learn the optimal strategies by deep reinforcement learning. For the third problem, we leverage Capsule Network to dynamically decide what and how much similar information need be transferred from each item, which can capture the user’s preference. Experiments on real datasets indicate that CapDRL outperforms state-of-the-art methods, validating the effectiveness of our approach on the recommender system. In addition, we explore the effects of different features on the proposed model.