Studying Movie Personalization Recommendation Based On SVR-CF And User Portrait Fusion

Watching movies is an important and necessary activity for people's daily leisure.The development of the film industry is in full swing and the number of movies has increased dramatically.The rapid development of the Internet has brought convenience to people's access to information,but it has also increase the number of movies on major video sites,causing people to be confused when faced with dazzling video resources.The current powerful tool for solving this problem is a recommendation system with information filtering and data mining.The most widely used technology in the movie recommendation system is collaborative filtering,which uses the past behavior or opinion of the existing user group to predict the movie most likely to be enjoyed by the current user.However,on the one hand,collaborative filtering does not take into account the content of the movie,thus ignoring the user's preference for the content of the movie;on the other hand,the input data for collaborative filtering is a user-movie rating matrix,and therefore it is vulnerable to the sparseness of user rating matrix.In view of the above problems,this paper proposes a hybrid recommendation algorithm based on SVR-CF and user portrait fusion.The main content of this paper is as follows:?1?Collaborative filtering ignores the user's preference for movie content and introduces user portrait technology.Firstly,the user history data is extracted to analyze the information that can represent the user's interest features.Then the user is modeled using the vector space model,in order to improve the recommendation efficiency at the same time,the K-means algorithm is used to cluster the user's portraits.Finally,the user's portraits are integrated into collaborative filtering to form a recommendation algorithm that combines collaborative filtering with user portraits.?2?For the collaborative filtering technology is vulnerable to the sparsity of the scoring matrix,the SVR model is used to predict and populate the user scoring matrix.First,using the advantages of SVR's nonlinear mapping theory and the fitting of complex data to forecast and fill the vacancy in the initial user rating matrix,and get an optimized user rating matrix.Then use the collaborative filtering algorithm to calculate the initial recommendation results,and finally use the user's portrait to second screening to get the final recommendation results to improve the recommendation effect.?3?Empirical research.Using the movie dataset provided by MovieLens website to calculate on the recommended algorithm of this paper,the results show that:when the nearest neighbor K=30,the recommended performance of the improved algorithm in this paper is optimal;although at the expense of accuracy rate,that is to say,the accuracy of the algorithm is slightly lower,it has a great advantage in terms of recall rate,comprehensive F₁ value,and recommended system operating time.