Research On Chinese Sentiment Analysis Based On Deep Neural Network

Since the 21 st century,belong with the development of Internet technology,especially the rapid developement of mobile Internet,various types of network applications have been rapidly popularized and people have increasingly benefited from the convenient services brought by the Internet.The number of Internet users has grown explosively.The number of social media users,such as WeChat,QQ,Twitter and Weibo has counted hundreds of millions.Data show that Sina Weibo has reached 462 million monthly users in 2018.People express their feelings and opinions on various events through Weibo and WeChat's friends.These data and information can reflect the public opinion very intuitively.Text sentiment analysis is the discovery of emotional views(his,anger,sadness,happiness,positive,negative,etc.)expressed in textual data.Effectively using a large amount of user's viewpoint information on the Internet,through the study of the subjective sentiments contained in these viewpoint data,to discover the subjective sentiments and opinions expressed by users on specific issues or products,has significant research significance.This paper presents some innovative Chinese Sentiment Analysis methods by analyzing the characteristics of Chinese text data and combining deep neural network related technologies.Based on the Convolutional Neural Networks and the Recurrent Neural Network model,effective Chinese sentiment analysis methods are proposed.Specifically,this paper has mainly achieved the following research results:1)For the sentiment analysis of Chinese Internet comment texts,this paper proposes a sentiment classification model textEBRNN based on Word-Embedding and BiLSTM networks.This model first trains a large number of Chinese corpus through word embedding techniques,and carries out word vectors for Chinese texts.And then through the BiLSTM for further emotional feature extraction,and finally through the classification network to establish a two-class model for sentiment classification,compared with the traditional machine learning methods such as support vector machines and textCNN model and LSTM network This method achieves the better classification accuracy and proves the effectiveness of the method.2)For the problem of sentiment analysis of Chinese network comment texts,this paper further proposes a BiLSTM network model with attention mechanism.Based on the original network model,the model fully considers the weight distribution of each feature in the sequence.The accuracy of the model is improved and the validity of the model is proved.3)For the emotional analysis of Chinese Weibo,we designed and implemented a deep emotion multi-classification model using BiLSTM-CNN networks and and the fusion modelwith attention mechanism.The fusion model compares the LSTM model and the CNN model and the BiLSTM has achieved better multi-category results,which proves the effectiveness of the fusion models of the two networks on the multiple emotions of Chinese microblog emotions.4)In this thesis,the pre-training word vector of neural network language model is studied.In the experiment,the network models of pre-trained word vectors and non-pre-trained word vectors are compared.It is proved that the neural network language model is effective in extracting shallow text features.The neural network language model can better extract shallow text features in text analysis problems.