Sentiment Classification Of Chinese Lyrics Based On Convolutional Neural Networks Model

With the popularization of the network,the music market in China has been rapidly developing,and various online music websites provide users with a wide variety of music.In order to make users quickly select their favorite music,these websites have almost classified music in different perspectives such as emotion,scenes,themes,etc.Related music researchers are also trying to find an automatically optimal method to classify music.Among these perspectives,classifying music according to emotion is a hot research topic at present.However,since the emotion of music is a kind of complex subjective feeling,it has a certain degree of ambiguity.Therefore,there are two problems: Firstly,the method of classifying music by artificial is complicated,and it is difficult to avoid the subjective influence.Secondly,the features of music such as audio and rhythm are relatively abstract and difficult to extract,which also brings certain difficulties to the research of automatic by computer.In order to solve the above problems,an emotional classification method for Chinese lyrics based on convolutional neural network is proposed in this paper.The method mainly involves two processes: the acquisition and representation of lyrics' feature items and the classification of Chinese music.Therefore,this paper firstly classifies the music of five online music websites by using truth table method and gives specific emotional labels.Secondly,analyzing the features of lyrics and the frequency of words,the types of words and the number of rows are counted as the reference elements to determine the number of feature items.For reference,calculating the TF-IDF value corresponding to the lyrics and using this as a criterion to compensate for the feature items and applying the skip-gram model of word2 vec to represent the extracted feature items as feature vectors,the last one of which is replaced by the TF-IDF value corresponding to the feature item.This method not only considers the characteristics of short lyrics and high repetition rate,but also avoids the problem of inaccurate classification caused by backfilling feature items.Finally,the matrix formed by the lyrics' corresponding feature vectors is used as the input of the convolutional neural network,and the lyric data sets are trained and tested by designing the layers of the convolutional neural network to achieve a better classification of lyric emotions.This paper collects a large amount of music data on the online music websites and summarizes the features of the lyrics.It statistically analyzes the number of words,the types of words,and the number of rows.It is sufficient after many training experiments and comparative analysis of the emotional classification results obtained by different methods.The accuracy and effectiveness of the proposed method are proved.At the same time,it will help in the future's research on text features' extraction and sentiment classification.