Research Of Text Sentiment Analysis Based On Deep Learning

With the Internet era,there are emerging platforms such as news portals,forums,and social networking sites.Emerging platforms have become an important platform for people to access important information.The data generated by the emerging platforms is mainly text,and these textual information are of great value.Text sentiment analysis uses algorithms to mine emotional tendency information in textual information.Text sentiment analysis is divided into three ways: based on dictionary text sentiment classification,based on machine learning text sentiment classification,and based on deep learning text sentiment classification.This paper divides text sentiment analysis into: long text sentiment analysis and short text sentiment analysis.A self-attention convolutional neural network model and a two-layer self-attention extended convolutional neural network are proposed for the above two sentiment analyses.The relevant research is as follows:(1)Self-focusing convolutional neural network model.The model solves the short text sentiment analysis,firstly preprocessing the short text,mainly completing the text segmentation and removing the stop words.The words that complete the pre-processing work are entered into the CBOW model,and the words are converted into word vectors.The vectorized words are extracted from the features of the text through the convolutional neural network,and then the information of the upper and lower words is merged through the bidirectional LSTM layer,and the self-attention mechanism is introduced to highlight the weight of the emotional words,thereby improving the accuracy of the text classification.The model is validated by public datasets,indicating the short text classification effect of the model.(2)Double-layer self-attention extended convolutional neural network model.The model solves long-text sentiment analysis,first long-term text pre-processing work,and also completes the operation of text segmentation and removal of stop words.The long text sequence data is long,and the parallel convolution kernel method is used to increase the receptive field and improve the range of extracting local features.After extracting the completed word vector features,the information of the upper and lower words is fused through the two-way GRU network,and the weight of the emotional words is emphasized to introduce the self-attention mechanism.The words of the same sentence are spliced together,the information of the upper and lower sentences is fused through the two-way GRU network,the weight of the emotional sentences is highlighted,and a self-attention mechanism is introduced.The model is validated by public datasets,indicating the long text classification effect of the model.(3)The application of text emotion in real data sets.The crawler program is used to obtain relevant information about Huawei on microblog.The information is used as short text data set,and news about the United States is obtained in today's headlines.The news is used as long text data set.The two models proposed in this paper are used to classify the text emotionally.The experimental results show that the model proposed in this paper is also effective and advanced in real data set.