A music emotion classification method based on SVM active learning

Abstract

The invention discloses an improved SVM (Support Vector Machine) active learning method, which relates to the cross field of music classification and information processing technology in machine learning, and also relates to an SVM active learning classification method oriented to large-scale training data. By adding the samples with the largest amount of information to the training set, the costof labeling samples can be greatly reduced by multiple iterations. To evaluate the performance of the classifier, the experiment comprises four kinds of music emotion categories: excited, angry, sad and relaxed, classifies music samples, and verifies the effectiveness of the SVM active learning method proposed by the invention from two aspects of the convergence speed of the classification accuracy rate and the number of samples to be labeled under the same accuracy rate.

Description

technical field [0001] The invention relates to the technical field of machine learning and information processing, in particular to a music emotion classification method based on SVM active learning. Background technique [0002] With the rapid development of Internet technology and multimedia information technology in recent years, the amount of information loaded on the network has become larger and larger, including the rapidly growing number of digital music. As one of the main ways of mass entertainment, music not only provides the public with It is an auditory feast, and the emotional connotation brought by music also greatly enriches the emotional space of the public. With the emergence of a large number of singers and the release of massive albums and online songs, and influenced by the diversification of world cultural development, various music styles have also emerged. In order to meet people's different preferences and emotional needs And quickly find the songs...

AI Technical Summary

Problems solved by technology

[0004] At present, among many research methods, most of the various music automatic classification methods that have been proposed belong to the traditional machine learning method, that is, the passive supervised learning method is adopted, that is, the training samples are manually marked by the user, and then the classification is obtained through training. Finally, the unknown sample data is classified. The classification effect of this passive learning method requires a large number of labeled training samples to participate in the training to be guaranteed, but the manpower and time spent on labeling these training samples are huge. For unknown samples Classification labeling also requires the labeler to have professional knowledge in the corresponding field

Images