基于音乐内容和歌词的音乐情感分类研究

摘要

According to the music emotion classification,an approach of multi-modal music emotion category combining music content and lyrics is proposed to compensate for lack of the single modal music emotion classification method that only uses music content for classification. Mainly discuss how to use lyrics for music emotion classification and combine music lyrics and content to improve the clas-sification accuracy. Using feature selection algorithm based on CHI and quadratic dimension reduction method based on Latent Semantic Analysis ( LSA) effectively improves the efficiency of text classification. For multi-modal fusion problem,propose an improved LFSM fusion method based on the traditional LFSM fusion method,and verify its feasibility through some experiments and compare the im-proved LFSM fusion method with the others. The results show that the accuracy of the improved method is highest,reaching 79. 51%, that verify the effectiveness of the method.%针对音乐情感分类问题，为了弥补仅仅利用音乐内容进行音乐情感分类的单一模态分类方法的不足，文中提出了结合音乐内容和歌词的多模态音乐情感分类的方法。主要探讨了如何利用歌词对音乐进行情感分类以及结合歌词和音乐内容以达到提高分类准确率的效果。对歌词进行特征选择时，分别利用CHI特征选择算法和潜在语义分析( LSA)对歌词进行降维处理，有效去除了噪声，提高了分类效率。针对多模态融合问题，在传统的LFSM融合方法的基础上，提出了改进的LFSM融合方法，并通过实验验证了该方法的可行性；同时将该方法与其他传统的融合方法的分类效果进行了比较。结果表明，改进的LFSM融合方法的分类准确率最高，达到了79.51%，验证了该方法的有效性。