A Research On Film Recommendation System Based On Combination Decision

With the rapid development of the Internet,people of the 21 st century have already stepped into an era with overloaded information from the age of information shortage.Among all platforms of video resources on the Internet,the traditional SE(search engine)needs to acquire information through keywords.This on-line technique,which highly relies on the preciseness of the keywords,may help people find their answer;however,there will be a large amount of optional information,which leads the users spending much time on those results that even some of them are non-related.Thus,the keywordsrelated technique has rarely met the needs of individuals.For the personalized-network era of the 21 st century,the system platform is required to actively analyze its users' behavior and to find out what are the users interested in as well as to connect the aimed information resources with its clients.Therefore,the recommender system is coming into effect.Firstly,throughout literatures at home and abroad,the author is aware of that the early recommendation technique was designed to divide neighbors by using the user's scoring of the items,to obtain the similarity of the scoring matrix,and to complete the recommendation based on the similar ideas of his neighbors.However,as the number of movies increasing,users only score a small amount of movies.Thus,the established scoring matrix is sparse which has eventually led to the decrease of the accuracy of recommendation.Therefore,this paper is written to study on movie recommendation through an idea of portfolio strategy.Secondly,the paper introduces a few recommendation techniques widely used at present,including the research content and applications.For the poor performance of recommendation toward sparse data,an improved combined recommendation algorithm model based on content and collaborative filtering is designed by using MovieLens 20 m data set provided by GroupLens.The center of the algorithm is to combine the features of a movie and the scoring system of the users,and figuring out the score of the same movie that each user scores toward different features.Finally,according to the project characteristics that each user likes,the users who like the same features are clustered.When making the recommendation,the first step is to analyse the user's preferred feature type through data and then clustering the same users into one group and calculating the similarity between the user and other users in the group.Finally,the algorithm finishes the recommendation in terms of the movies that the neighboring users likes.At last,the paper makes an introduction to the collaborative filtering recommendation based on SVD(Singular Value Decomposition).The collaborative filtering recommendation is to decompose a matrix into two orthogonal matrices and a diagonal matrix through the way of SVD.The original data set can be represented by the data set decomposed by SVD.After eliminating the redundant data with linear correlation,the unnecessary attributes and features can be reduced in data processing.Thus,the recommendation can be completed by using the collaborative filtering idea.What's more,the prediction accuracy is compared by RMSE(root mean square error),MSE(mean square error),and MAE(absolute mean square error)comparison algorithm.which indicates that the combined algorithm is superior to the traditional recommendation methods.