Sentiment Analysis Of Chinese Commentary Text Based On Deep Learning And Attention Mechanism

With the increasing development of the Internet and the continuous improvement of the functions of mobile terminals,people communicate and interact with the outside world more frequently through many different social networking platforms and website APPs.People can comment on the news,things,goods they buy and so on.These comments contain information about people’s emotional tendencies towards different things.It is of great significance for individuals,enterprises,society and government departments to dig out the emotions and classify them.Chinese comment sentiment analysis is more difficult than English comment sentiment analysis,which is a hot research field in natural language processing.Traditional sentiment analysis methods based on sentiment dictionary and machine learning have been unable to meet the needs of sentiment analysis for the massive review data.Therefore,this paper takes the comments of Dianping and the comments of Jing dong as the research objects,studies the sentiment analysis method based on deep learning and attentional mechanism,and excavates the emotional features in the comment information,so as to judge the emotional polarity of the comment information.First of all,traditional SVM and deep learning LSTM,GRU,TCN,BILSTM,BIGRU,Bit CN and other sentiment analysis models were studied,and comparative experimental analysis was carried out to conclude that the model based on deep learning is better than the traditional model based on machine learning.Secondly,in order to improve the effect of comment text sentiment analysis on the basis of the basic model,we conduct research on attentional mechanisms,including self-attentional mechanism(SA)and multi-headed attentional mechanism(MA).Based on the research of BiTCN sentiment analysis model,the BiTCN-Attantion sentiment analysis model is proposed.On the basis of BiTCN model,self-attentional mechanism and multi-headed attentional mechanism are added respectively,so as to focus on multiple semantic centers and understand the full text more accurately and efficiently,aiming to improve the accuracy of the model.Finally,the BiTCN sentiment analysis model based on attention mechanism proposed in this paper is verified by experiments.By selecting two Chinese datasets and preprocessing them,such as sentiment labeling,text cleaning and Chinese word segmentation,the corpus of comment text in this paper is obtained.Word2 vec model is used to vectorize the preprocessed comment text.Through comparative experiments,specific values of different parameters in the TCN network model are determined,such as the size of the convolution kernel,the number of convolution layers,the number of iterations,the optimizer,the expansion factor,etc.,and the parameters that can perform better in the data set selected in this paper are selected.By comparing the experimental effects of the deep learning model BiTCN with the BiTCN-Attention model,the following conclusions can be drawn:The deep learning model with the added attentional mechanism is better than the relatively simple deep learning model,because the attentional mechanism can improve the weight of affective words,thus improving the accuracy of feature extraction;The self-attentional mechanism is superior to the multi-head self-attentional mechanism in the data set of this paper.According to the experimental data,the accuracy of the model of BiTCN-SA in the commodity review data set of Jingdong is 3.96% and 2.41% higher than that of BiTCN and BiTCN-MA;in the comment data set of Dianping,the model accuracy of BiTCN-SA is 4.62% and 3.49% higher than that of BiTCN and BiTCN-MA,respectively.Thus,the validity of the proposed model is verified.