Acquiring Mood Information from Songs in Large Music Database
Yi Liu, Yue Gao · 2009
Automatic mood information acquiring from music data is an important topic of music retrieval area. In this paper, we try to find the strongest emotional expression of the song in large music databases. By analyzing hundreds of credible reviews from website, a 7 keywords mood model is constructed. 217 songs were collected in our dataset. Every song was divided into several 10s-long segments and our dataset containing 5929 music clips. We used Gaussian Mixture Model (GMM) and Support Vector Machine (SVM) as classifier and four feature selection algorithms to do mood classification experiments. A post-processing method was presented to find the strongest mood expression of each song. From the experiment result, we can see that SVM is the best classifier for mood classification, and Active Selection algorithm can remove weak features effectively. Using SVM classifier, the classification accuracy can achieves 83.33% with 40 features by using active selection algorithm, and 85.42% with 84 features which selected by ReliefF.