Research And Application Of SVD++ Algorithm In Video Recommendation System

The widespread use of the Internet has promoted the development of education information.Online learning resources have sprung up like mushrooms after a rain.The huge number of learning resources has increased the pressure of users to choose.How to help users accurately filter out the resources that meet their individual needs from the massive network resources has become the core of the recommendation system research,and a large number of scholars have invested in the research and construction of the personalized recommendation of educational resources.This paper designs and develops an office online video learning recommendation system based on the improved SVD++ algorithm.Aiming at the data sparseness problem in the SVD++ model,a method of using tag information to describe videos and user preferences is proposed,which makes the calculation of similarity between users more reasonable,improves the prediction accuracy,and recommends videos that users are really interested in.The research work of the thesis mainly includes the following four aspects:1.Use clustering algorithm to preprocess the data.In order to reduce the memory consumption of each iteration of the algorithm,users and videos are clustered separately based on the K-means clustering algorithm,and the global user-rating matrix is split into small matrices with similar characteristics according to user similarity and video similarity.Use it as the input for subsequent model training,reducing the computational complexity of the recommendation algorithm.2.Propose a method to solve the problem of system data sparseness by using video tags.Set knowledge point labels for videos,use TF-IDF weighting technology to calculate the user's preference for tags,mine user preferences,and then build a user preference similarity matrix,which solves the data sparseness problem in the userscoring matrix.3.Propose an improved SVD++ algorithm that integrates time information and user preference similarity matrix.Introduce the time half-life and information retention period into the forgetting function,and use the new time weight function to improve the user-label preferences and static user factor preferences in the model,and finally get the optimized SVD++ algorithm.The RMSE index is used to verify the accuracy of this algorithm on the preprocessed ml-latest data collection.Experiments show that the optimized SVD++ algorithm can improve the accuracy of the recommendation.4.Implement an office video learning recommendation system.The system uses the Pycharm integrated development environment and Mysql8.0 platform to build,uses Vue technology and the MVT design pattern of the Django framework for development,and realizes the functions of user registration and login,uploading videos,and video classification retrieval.The test verifies that the system can use the improved SVD++algorithm to make personalized learning video recommendations for users.