Research On Emotion Classification Of Chinese Short Texts Based On Deep Learning

At present,most of the texts on the Internet are short texts,such as WeChat and weibo.The short text is characterized by sparse features,irregular expression and diverse expression of emotions.The traditional methods and techniques for emotion classification of long text are not suitable for emotion classification of short text.Therefore,this paper applies the deep learning method,which has made a great breakthrough in the field of image,to the Chinese short text emotion classification,and designs a technical scheme for Chinese short text emotion classification based on lstm-cnn.This paper preprocesses the data according to the characteristics of Chinese text,and proves the effectiveness of the scheme through a large number of experiments.The main work of this paper is reflected in the following aspects:Firstly,this paper constructs a corpus of emotion classification of Chinese short texts.Since there is no recognized Chinese emotion multi-classification corpus with a large amount of data at present,this paper identifies two positive emotions of joy and praise and four negative emotions of disappointment,anger,hatred and fear.On this basis,this paper collects and collates relevant data.A total of more than 20,000 emotional corpora with emotional labels have been screened and sorted out,which constitute the emotional classification corpus.Secondly,this paper preprocesses the corpus of emotion classification.The preprocessing of corpus mainly includes two aspects: Chinese word segmentation and word vector training.In this paper,jieba word segmentation kit is used for word segmentation of Chinese corpus.Then,the skip-gram model in word2 vec was used to train the corpus into a word vector for classification.Finally,this paper proposes a Chinese text emotion classification model based on lstm-cnn.This model combines the methods of LSTM and CNN in deep learning and can be divided into four parts: firstly,the semantic features of the text are extracted through the LSTM layer.the attention mechanism is introduced to pay special attention to the position of important features,and the feature vectors are vertically stacked into a two-dimensional matrix.Secondly,the twodimensional matrix is input into the convolution layer,and three convolution kernels of different sizes are set in the convolution layer to check the characteristics of different granularity for preliminary extraction.Thirdly,the maximum pooling method is adopted in the pooling layer to extract the features of the convolution layer.Fourthly,the extracted features are integrated through the full connection layer.Finally,input the obtained features into the softmax classifier to complete the classification task.By comparing the experiment has proved that compared with other classification model,this article puts forward the classification of the model has better accuracy.