Research And Implementation On Sentiment Classification Of Chinese Folk Songs

The emotional analysis of Chinese folk songs has been studied in many fields.In the field of musicology,scholars' emotional research on folk songs mainly focuses on the emotional expression of specific regions or specific folk songs;in the computer field,it is mainly based on emotional classification of music audio or lyrics.This thesis takes folk song lyrics as the research object,uses Word2 vec word vector to extract text features,and uses convolutional neural network algorithm to realize folk song emotion classification.The specific research contents include: firstly,the Chinese folk songs and songs corpus are built.The corpus includes a total of 1082 Chinese folk song lyrics,and the usage data is expanded to 3615,which are marked as“warm and cheerful”,“sad sadness” and “other categories”.Secondly,pre-processing the lyrics text,including de-stopping words,removing song names,lyrics,composition,singers,singing time and other useless repetitive information and word segmentation.In the process of word segmentation,this thesis selects three word segmentation tools for comparison,and selects the jieba word segmentation tool according to the experimental results.Then use the one-hot encoding vectorization representation of the text,and Then use the one-hot encoding vectorization representation of the text,and then use Word2 vec to extract the features of the lyric text.Combined with the corpus characteristics of the corpus,compared with the traditional machine learning classification algorithm Naive Bayes,support vector machine,K-nearest neighbor,the convolutional neural network is finally selected as the classifier of folk song lyrics emotion,and different sizes are proposed.The CNN-1C model of the convolution kernel was tested.Comparing the experimental results,it is concluded that the CNN-1C model based on word-level features is more suitable for the classification of the folk songs and songs.The F1 value of the classification results can reach 76.7%.Finally,a summary of the work was carried out and further work was proposed.