Research On Chinese Music Emotion Classification Based On Lyrics And Comments

In the era of streaming media,the classification of songs is one of the main methods to manage the massive songs,realize precise personalized selection and recommendation and enhance the experience of audience,and the common classification elements include genre,emotion,musical instrument,language,theme,etc.As an important moving way and communication medium for human beings,songs carry rich emotional information.Recently,four song emotion classification methods have emerged,based on content(melody and tune),lyrics,the blending of content and lyrics,and the social tags.However,due to the different styles of modern songs and the relatively implicit expression of Chinese lyrics,these methods do not perform well in the emotional classification of Chinese songs.In response to this shortcoming,the listener's perceived and experiential emotion have been considered comprehensively and a new emotion classification method of Chinese songs based on comments and lyrics is proposed in this dissertation.The main work of this essay is as follows:Firstly,a corpus of sentiment classification for Chinese songs is constructed.At the beginning,the lyrics and comment corpus is obtained through web crawler.Aiming at the difference between lyrics and comments,then,the different corpus cleaning programs and stop word lists are constructed,and through comparative experiments,this essay chooses the different word segmentation tools.A pre-trained BERT language model is used to filter out valid information from the data,and the effectiveness of the sifting process is verified by statistical and experimental methods.Secondly,a sentiment classification method for Chinese songs based on comments and lyrics is proposed.Based on the traditional lyrics-based method,this method introduces the listener comment,which is the element of the listener's intuitive perception of emotion and experience emotion expression,uses the early fusion by feature concatenation method and late fusion by linear combination method to merge the lyrics and comment features,and uses SVM,KNN,CNN and LSTM four classic classification algorithms for method implementation.Thirdly,compared with traditional lyrics-based methods through experiments,on the four classification models of SVM,KNN,CNN and LSTM,the effect of the proposed method is better than the lyrics-based song sentiment classification method.The accuracy of test set classification is improved by 3.4%,which verifies the effectiveness of the proposed method.