Research On Network Public Opinion Analysis Method Based On Deep Learning

With the continuous development of science and technology,the impact of the Internet on people's daily life is also increasing.People are more and more inclined to publish their views and opinions on some things on social networks.Public opinion analysis can quickly judge people's emotional tendency to major events and objectively express the direction of social public opinion,which is conducive to the timely and accurate management and guidance of public opinion information by specific competent departments and government agencies,so as to maintain social harmony and security.Most of the public opinion information comes from the short text comment information.The text is separated from the written language,the structure becomes more concise and lack of standardization,which often causes certain difficulty to the text feature extraction.Traditional affective analysis methods often rely on affective dictionaries and feature extraction.With the continuous update and iteration of Internet culture and data volume,a large number of people are required to update affective dictionaries.Otherwise,semantic features will be lost and classification will be inaccurate.This paper is based on the analysis of attention mechanism and deep learning technology,an han-clstm model is proposed to mine the deep semantic features of text,which can accurately judge its emotional tendency.The main research work of this paper includes the following parts:According to the characteristics of CNN and LSTM in text processing,CNN can extract the local features of text better,LSTM can retain the historical information of text,and extract the global features of sequence effectively.In order to make the extracted feature semantic information more comprehensive,the two are combined to form clstm model.The clstm model with the optimal parameters is obtained by comparing several experiments with several groups of model parameters,which improves the classification performance compared with the traditional CNN model and LSTM model.Aiming at the problem that clstm model can't effectively extract the hierarchical relationship between sentences and doesn't consider the distribution of weight to feature vectors,a hierarchical attention mechanism is introduced to optimize clstm model,and an han-clstm model is proposed to classify the emotional tendency of texts.This model assigns different weights to different word level feature vectors and sentence level feature vectors,so that the model can In the process of calculation,more attention is paid to the eigenvectors which have influence on the classification results.At last,the experiment of nlpcc data set shows that the performance of the improved han-clstm model is better than that of clstm model.The experiment results show that the accuracy,recall rate and F1 value of the performance evaluation index of the model are higher than those of other classification models.