Research And Application Of Video Recommendation Algorithm Based On Neural Network

With the continuous expansion of the Internet and the subsequent era of big data,various kinds of information on the Internet has become more and more,resulting in information explosion and information overload,making users unable to distinguish and obtain the desired information in a short time.The function of the recommendation system is to provide users with key information.Through deep algorithm mining of users' personal information and historical data,the recommendation system can get users' interest demands,and finally give a personalized recommendation list.The core idea of collaborative filtering algorithm is to find users with similar behaviors,which is a group with common needs and hobbies.The algorithm is more inclined to pay attention to users' historical behaviors and is not affected by new projects,which has better recommendation accuracy.In this paper,contentr-based and user-based collaborative filtering algorithms are analyzed and compared,and clustering algorithms are added in the process of practical optimization.The recommendation accuracy of collaborative filtering algorithm is not high,Therefore,the similarity formula is optimized according to the calculation of time interest degree,and before calculating user similarity,using the clustering algorithm to cluster,users will have similar behavior in aggregated into a class.In this way,the time to calculate the similarity between users can be greatly shortened in the experimental stage.The experiment shows that compared with the single recommendation algorithm,the hybrid recommendation algorithm is further improved on the criteria of accuracy and recall.At the same time,self-coding neural network is a kind of unsupervised learning deep network model with strong hidden layer feature extraction ability.Therefore,a hybrid recommendation algorithm based on stack de-noising self-encoder is proposed.first of all,the data set of user rating data is used for training the network input,to learn the implied character encoding of the project,and then use PCA algorithm to reduce the dimension of the project properties,finally,combined with the user data to calculate the similarity after the optimization,get the final TOP-N recommended list.Experimental data use the public Movielens data set,70% for training sets and 30% for test sets.Experiments show that the new algorithm can solve the sparsity problem of the scoring matrix to some extent,because the depth model can extract user characteristics effectively,and the recall rate and accuracy of the recommended results are also improved.Therefore,the hybrid optimization of neural network model and collaborative filtering can greatly promote the development of recommendation system.