基于位置权重的歌词情感分析

Research on Lyric Sentiment Analysis Based on Position Weight

结合目前从音频和歌词角度对歌曲情感分析的研究以及歌词文本独有特点,提出一种基于文本标题与位置权重相结合的歌词情感分析方法。该方法考虑到出现在不同位置的特征词权值对于歌词分类的影响程度,采用层次分析法计算文本标题、歌词文本前、中、后不同位置特征词的位置权重。通过朴素贝叶斯、最大熵模型、支持向量机等不同分类器的训练实现歌曲快乐、伤感、安静、激昂4种情感分类。实验表明,加入文本标题与位置权重后的最优F1值相比之前提升了4个百分点,证明了该方法在提升歌词情感分类性能方面的有效性。

Combining the current research on emotional analysis of songs from the perspective of audio and lyrics and the unique characteristics of lyrics text,we propose a method on lyrics sentiment analysis which is based on the text title and position weight.The algorithm takes into account of the influence degree of the feature words appearing in different positions on the classification of the songs.AHP is used to calculate the position weights of the text title,the front,middle and back of the text in different positions and trained by the NB,ME,SVM different classifier,songs are finally divided into four emotional classification including happiness,sadness,quietness and passion.Experiments show that the optimal F1value after adding the text title and position weight is increased by4percentage points,which proves the effectiveness of this method in enhancing the performance of lyric sentiment classification.