Application And Implementation Of Movie Recommendation Algorithm Based On Reinforcement Learning

The demands of internet users on movies are increasing,with the rapid development of mobile Internet and the increase of network speed.The number of movies on movie websites was continuously increasing,which made it difficult for users to find movies of interest.The evolving recommendation algorithm solves the information overload problem of the movie to a certain extent,which accurately recommended movies for the users.Most of the offline models with a fixed recommendation strategy can't perceive the changes of user interest.High similarity and poor diversity of recommendation cannot meet the users' dynamic interest.The value of the recommendation system is not fully utilized.Therefore,it is valuable to design a recommendation algorithm with diversity and applied to the film recommendation system.A film recommendation algorithm based on reinforcement learning is proposed and applied to the movie recommendation prototype system,according to the user's high degree of sparseness on the film's scoring behavior and the dynamic change of the user's viewing movie interest over time.The TopN recommendation experiment on the MovieLens dataset shows that the film recommendation model based on reinforcement learning can improve the overall diversity while ensuring recommendation accuracy rate,compared with the traditional collaborative filtering model and deep learning model,when the number of recommended movies is large.The Markov decision process based on reinforcement learning is modeled to simulate recommendation system recommending different movies for the user according to the user's state.The optimal candidate movie is selected by the value function,and the ?-greedy random strategy is introduced as the final recommendation result to further improve the diversity.The deep Q-network was trained with the experience replay technology,which saved the user's viewing movie status,movie recommendation action,reward and new user status as samples to the experience pool,and randomly take some samples from the experience pool for supervised learning.The user's dynamic interest was taken into the movie recommendation process with reinforcement learning based algorithms,which helps to improve the diversity of recommendation results.The TopN recommendation experiment on the MovieLens dataset shows that when the number of recommended movies is large,the reinforcement learning based film recommendation algorithm can guarantee the accuracy of recommendation,and the overall diversity is improved compared with the hybrid algorithm of collaborative filtering and linUCB.