Music Recommendation System Based On Convolutional Neural Network

With the development of technology,the music library becomes larger and richer,which makes it difficult for users to find their favorite music.Personalized music recommendation system is proposed to solve this problem while there exists many kinds of shortcomings necessary to be improved.Convolutional neural network is an efficient deep learning method,which is widely used in many fields such as image recognition,natural language processing achieving better performance than traditional techniques.According to what is mentioned above,this thesis proposes music recommendation system based on convolutional neural network.The music recommendation system proposed in this paper includes user modeling module,audio feature extraction module and recommendation algorithm module,which firstly constructs the user preference model by matrix factorization of implicit semantic model after collecting the historical behavior of listeners,and then extractes the Mel spectrum which can represent the audio from the music that has been preprocessed,and next trains convolutional neural network to obtain the regression model for predicting the potential features of music which can project users and music into a shared hidden space,and finally recommends the top N music for target users on the basis of the similarity between user preference features and music potential features.Systematical test is used to verify the performance of the proposed music recommendation system.This thesis constructes the experiment datasets for training and testing model,designs the network model used in this experiment on the basis of the typical convolutional neural network model,and optimizes the model.In this paper,root mean square error,accuracy,recall rate and F1 value are used to evaluate the system from the aspects of prediction score accuracy and recommendation list accuracy.The results show that the proposed algorithm is feasible and effective,which makes the best use of the advantage of deep neural network to automatically acquire higher-level music features in audio and incorporates the historical behavior of users to music compared with the traditional music recommendation algorithm,improving the problem of cold start existed in the recommendation system.