PROJECT TITLE :

Social Attentive Deep Q-networks for Recommender Systems

ABSTRACT:

The purpose of recommender systems is to actively and accurately present users with items that could be interesting to them (products, information or services). Deep Reinforcement Learning has been successfully applied to recommender systems, but it still suffers heavily from data sparsity and cold-start when it is used in real-world tasks. In this study, we propose an efficient method for dealing with problems of this nature by utilizing the pervasive Social Networks that exist among users in the process of estimating action-values (Q). To be more specific, we develop a Social Attentive Deep Q-network (SADQN) to approximate the optimal action-value function based on the preferences of both individual users and social neighbors. We do this by successfully utilizing a social attention layer to model the influence between them. This allows us to approximate the optimal action-value function. In addition, we propose an improved version of SADQN, which we will refer to as SADQN++, to model the complicated and diverse trade-offs between personal preferences and social influence for all users involved. This will make the agent more capable and flexible in terms of learning the policies that are optimal. The experimental results on real-world datasets demonstrate that the proposed SADQNs perform noticeably better than the state-of-the-art Deep Reinforcement Learning agents, with a reasonable increase in the amount of computation required.