Music similarity measurement and recommendation system using convolutional neural networks

Abstract

Due to the massive and growing volume of music on the internet, and the lack of proper management on this massive volume, similarity and music recommendation systems have been designed. The music similarity system, which is the basis of the recommendation system, can automatically generate a user's playlist according to the similar features of each piece of music. In this paper, we designed a desirable music genre classification using convolutional neural network for extracting high-level features from intermediate networks layers. For similarity measurement, we considered cosine similarity and Euclidean distances between feature vectors. We applied this automatic recommendation system on three databases with different genres and showed that the recommender achieves significant accuracy in 10-Best results.