音乐和歌词融合的歌曲情感分类研究

Research on Music Mood Classification Integrating Audio and Lyrics

为更好地对歌词进行情感分类,提出一种改进的基于类间差别的CHI特征选择方法。该方法可单独用于歌词情感特征提取,将选取的特征应用于支持向量机分类器中,融合音频特征与利用改进CHI方法选择的歌词特征对歌曲进行情感分类。实验结果表明,融合后的特征可以取得比任何单一种类特征更好的分类效果。

In view of the distinct levels of the association of a term with different classes, an improved difference-based CHI method is proposed to extract discriminative affective words form lyrics text. Support Vector Macbine（SVM） classifier is constructed to apply the selected features, obviously increasing lyric sentiment classification performance. The lyric features selected by the improved method are combined with audio features. Experimental results verify the efficiency of this fusion.