Dynamic Movie Recommendation Algorithm Based on Tag Genome and Deep Reinforcement Learning in Large Discrete Action Spaces

Abstract

In this paper we proposed a recommendation algorithm which is based on deep reinforcement learning to define its architecture and related parameters According to different users click on different movies and their ratings personalized recommendations for users In addition to combining the characteristics of different recommendation algorithms such as the advantages of collaborative filtering and content-based recommendation It is improved for problems in large discrete action spaces of reinforcement learning hoping to improve system performance and make good recommend Through the experimental results we found that Comparison with DQN algorithm in a large number of actions our method can converge at an early epoch and increase the speed by more than 20 times Comparison with other commonly used recommendation algorithm shows that it performs better in multiple indicators and can dynamically update the model The method can give good recommendations even for users who have never seen the system at all There is no overfitting problem in the verification and test dataset proving the universality of this model

Date of Award	2020
Original language	English
Supervisor	Chin-Feng Lai (Supervisor)