Deep Learning-based Music Data Analysis And Personalized Recommendation

After hundreds of years of development in the domestic and international digital music market,the total number of recorded music works has reached a considerable level.Faced with such a large number of music works,how can users listen to their favorite music works more conveniently and efficiently? It is something that music platforms must consider,and it is also a research topic that researchers are very interested in.Regarding music data analysis and personalized recommendation algorithms,current research is mainly focused on the improvement of classic collaborative filtering recommendation,content recommendation,model recommendation and other traditional recommendation systems,while the problem of cold start of the recommendation system and the sparseness of the scoring matrix The nature has not been a good solution.The Deep Belief Neural Network(DBN-DNN)is a deep learning model nested by many restricted Boltzmann machines(RBM),with RBM processing sparsity matrix and real-time,at the same time,can produce better initialization parameters,greatly improving the training speed of the model.This paper uses a hidden semantic model to perform matrix decomposition on user ratings,extracts the user’s preference features for k hidden factors,and the weight of music works on these k hidden factors,and then considers that music itself also has categorizable information,through audio features The analysis method is to extract audio feature values ??from music fragments,and finally combine deep belief neural networks to predict users’ ratings of music.The first part gives a detailed introduction to the theoretical knowledge of collaborative filtering,audio feature analysis and deep belief neural networks;The second part gives all the algorithm steps related to preference extraction,feature extraction,and neural network training;the third part uses the data set to empirically analyze and optimize the algorithm as a whole,and explain the experimental results.Experiments show that the personalized recommendation algorithm combined with the deep belief neural network has good recommendation performance.For the scored samples,the recommended coefficient predicted by the model is less different than the real score.At the same time,the trained neural network model can be very good.To deal with the cold start problem in terms of items,for samples that users have not scored,the recommended error predicted by the model is within an acceptable range,so that the recommendation is no longer limited to the original user group and music library range,with better recommendation accuracy Degree and better extensibility.