Reinforcement Learning based Recommender System using Q-Learning and Deep Q-Learning

Abstract

The first part of this thesis concludes with an overall summary of the publications so far on the recommender system using reinforcement learning. We have performed a systematic review of the research studies published from 2010-2022. The second part of this thesis demonstrates Q-learning and Deep Q-learning applied in movie recommender systems using the Netflix prize data set. The limitations of traditional recommender systems, such as collaborative filtering, content-based filtering, and hybrid techniques, include data sparsity and cold start. To address these challenges, reinforcement learning methods for providing recommendations have become increasingly popular. We have built a model to improve the quality of recommendation by capturing the constantly changing and sequential structure of the recommendation issue and utilizing reinforcement learning to train prediction models (RL). As a case study, we construct this system using the well-known Netflix Prize data set and create a movie recommender system. We describe how these ideas may be simply changed for additional user-item interactions, as well as the outcomes and problems. Recommender systems are often utilized to provide customized content to users. We developed a novel algorithm that attempts to bridge several breakthroughs in the area, such as reinforcement learning methods. We demonstrated that quality and ongoing recommendations can be learned using the Netflix Prize data set. It is a dynamic system that improves on the static nature of previous recommender systems, and it is a versatile technique that can be adapted to many different applications that model and anticipate interactions between users and things.