Research And Implementation Of Recommendation Algorithm Based On Machine Learning

With the deepening of the Internet and the explosive growth of mobile Internet,the amount of resources accumulated on the network has increased geometrically,far beyond the ability of users to see and understand in time,which is called information overload.Personalized recommendation technology is an important way to effectively solve information overload problem,it can help users or systems to get information efficiently and meet users' personalized needs in time.This paper mainly research the movie recommendation algorithm based on the long and short time sequence network and the cold start method of music recommendation based on the convolution neural network.In the movie recommendation algorithm,we need to score the movie that users never see before.The conventional matrix decomposition method ignores the rule of user interest and the degree of movie spread over time.Therefore,this article in the training phase model by constructing vector and vector respectively to accept user input of the two film as LSTM neural network,feature capture users and movies in the temporal variation of the two sub LSTM neural network through the output compressed as another LSTM neural network input,the user rating of the film as the output of the model.The film scoring model based on LSTM was trained through the history of film scoring,and the effectiveness of the model was verified by comparison with SVD,PMF and RRN networks.In the research of music cold start solution,this paper proposes a music classification model based on convolution neural network to solve the algorithm of music cold start problem,and reduce the work of new manual music labeling system.Firstly music audio conversion into monochromatic spectrum by discrete Fu Liye change,then the spectral graph cut into uniform size,and finally the image input multi convolution music classification network is constructed,based on multi classification model to judge the convolution of music type of music,the new music recommendation to love this type of user.The experimental results show that the proposed music classifier based on convolutional neural network performs well in music and audio crawling in the shrimp community,and the accuracy of the music classifier based on cosine similarity is increased by 3%compared with the music classifier based on cosine similarity.