面向中文歌词的音乐情感分类方法

情感是音乐最重要的语义信息，音乐情感分类广泛应用于音乐检索，音乐推荐和音乐治疗等领域.传统的音乐情感分类大都是基于音频的，但基于现在的技术水平，很难从音频中提取出语义相关的音频特征.歌词文本中蕴含着一些情感信息，结合歌词进行音乐情感分类可以进一步提高分类性能.本文将面向中文歌词进行研究，构建一部合理的音乐情感词典是歌词情感分析的前提和基础，因此基于Word2Vec构建音乐领域的中文情感词典，并基于情感词加权和词性进行中文音乐情感分析.本文首先以VA情感模型为基础构建情感词表，采用Word2Vec中词语相似度计算的思想扩展情感词表，构建中文音乐情感词典，词典中包含每个词的情感类别和情感权值.然后，依照该词典获取情感词权值，构建基于TF-IDF （Term Frequency-Inverse Document Frequency）和词性的歌词文本的特征向量，最终实现音乐情感分类.实验结果表明所构建的音乐情感词典更适用于音乐领域，同时在构造特征向量时考虑词性的影响也可以提高准确率.

Classification of Musical Emotions Oriented to Chinese Lyrics

Emotion is the most important semantic information of music. Music emotional classification is widely used in music retrieval, music recommendation, and music therapy. Traditional music emotional classification is mostly based on audio. Nevertheless, based on the current technology level, it is difficult to extract semantic-related audio features from audio. There is some emotional information in the Lyric text, and music emotional classification is carried out with lyrics. This study focuses on Chinese lyrics and constructs a reasonable music emotion dictionary, which is the premise and foundation of lyric emotion analysis. Therefore, a Chinese emotion dictionary in music field is constructed based on Word2Vec, and Chinese music emotion analysis is carried out based on the weighting of emotional words and the part of speech. Firstly, this study constructs the emotional lyrics table based on the VA emotional model and adopts Word2Vec. The idea of word similarity calculation in Word2Vec extends the emotional vocabulary and constructs a Chinese music emotional dictionary, which contains the emotional categories and emotional weights of each word. Then, according to the dictionary, emotional words weights are obtained, and feature vectors of lyric texts based on TF-IDF (Term Frequency-Inverse Document Frequency) and lexical features are constructed. Finally, music emotional classification is realized. The constructed music emotion dictionary is more suitable for music field, and the accuracy can be improved by considering the influence of part of speech when constructing feature vectors.