The application of neural network with convolution algorithm in Western music recommendation practice

Abstract

With the rapid development of digital music, the number of Western music works is continuously increasing, which makes users find it difficult to spot their favorite music works consequently quickly. Therefore, the music recommendation algorithm is applied to recommend music works in a targeted manner based on prior user behaviors, which could reduce the fatigue of the users and improve the overall user experiences. The convolutional neural network (CNN) is applied to classify the commonly-seen types of Western music, including classical music, pop music, jazz music, and Hip–Hop and Rap music. Afterward, CNN is trained to explore the two activation functions and the two gradient descent methods, which are compared and analyzed in terms of their features and performances during the training. Then, the classification methods, which are based on the spectrum and the comprehensive feature frequency of spectrum and musical notes, respectively, are compared. Research results have shown that the accuracy rate of spectrum-based classification method is 96.5%, while that of the classification method based on the comprehensive feature frequency of spectrum and musical notes increases by 2%. Thus, the proposed music classification algorithm is significant to the extraction of music high-level semantic features, as well as the promotion of deep learning method in the field of musical signal analysis.