Research On Hybrid Recommendation Algorithm Based On Movie Rating

The problem of information overload and information trek in movie video websites is very obvious.As an effective method to solve such problems,the recommendation system has become a key research area for businessmen and researchers of various movie websites.The rating of movie is the most popular data that most researchers study,they usually start the verification performance experiment of recommendation algorithm on simple processing of the rating data,but the experimental result is almost same as on the unprocessed data.So this article proposes a novel user-based rating filtering method.Experimental results show that this new method can give full play to the performance of each recommendation algorithm.The recommended methods in the movie recommendation system can be divided into three categories: content-based filtering methods,collaborative filtering methods and hybrid methods of them.The most widely used and best performing method is the hybrid methods because they can achieve complementary of their strengths and weaknesses.This article uses the mixture method of improved based movie categories filtering method and singular value decomposition model in the collaborative filtering method for rating prediction,which based on the rating from movie website.The purpose of the article is to generate a predictive rating based on rating of movie dataset.On the selection of the data set,two publicly available real movie data sets hetrec2011-movielens-2k-v2 and ml-latest were selected for experiment.An off-line experiment method used 5 fold and 5 cross-validation in the movie recommendation process.Finally,the experiment used Root Mean Square Error and Mean Absolute Error as the evaluation indicators of the algorithm recommendation evaluation index.In the selection of rating data,this article proposes a novel user-based datafiltering method compared to the traditional method of randomly extracting or extracting rating of a specified user(users' ratings are greater than a certain threshold).Experiments show that the rating processed by this method can greatly reduce Root Mean Square Error and Mean Absolute Error of the rating prediction results of each recommendation algorithm.The content-based filtering method aims to analyze the characteristics of the users and the items to establish a similarity profile,then recommend similar items to the users based on the similarity file or predict that the users will give similar ratings to similar movies.In the movie data set,since the categories of the movies have a great influence on the similarity of the movie and the analysis of the user's preference,this article uses the movie categories attribute to establish the similarity file.The traditional movies categories-based filtering method is to directly apply the categories attribute to the analysis of the movies' similarity.This article proposes a novel filtering method based on the movie categories: using the movie categories to influence the user's existing ratings.The ratings incorporates the categories of movie into the user's rating,then using singular value decomposition model to make rating prediction based on the new rating.The experiment proves the effectiveness of the proposed combined algorithm.