Reinforcement Learning based Recommender Systems: A Survey

Recommender systems (RSs) have become an inseparable part of our everyday lives. They help us find favorite items to purchase, friends on social networks, and movies watch. Traditionally, the recommendation problem was considered be a classification or prediction problem, but it is now widely agreed that formulating as sequential decision can better reflect user-system interaction. Therefore, formulated Markov process (MDP) solved by reinforcement learning (RL) algorithms. Unlike traditional methods, including collaborative filtering content-based filtering, RL able handle sequential, dynamic interaction take into account long-term user engagement. Although idea using for not new has been around about two decades, very practical, mainly because scalability problems However, trend emerged in field since introduction deep (DRL) , which made possible apply with large state action spaces. In this paper, survey based recommender (RLRSs) presented. Our aim present outlook provide reader fairly complete knowledge key concepts field. We first recognize illustrate RLRSs generally classified RL- DRL-based methods. Then, we propose RLRS framework four components, i.e., representation, policy optimization, reward formulation, environment building, algorithms accordingly. highlight emerging topics depict important trends various graphs tables. Finally, discuss aspects challenges addressed future.