Research On The Application Of Network Sentiment Influence Model In Dissemination Of Public Opinion On Emergencies

In recent years,China’s frequent emergencies have become a prominent problem that can not be ignored in contemporary society.When an emergency happens unpredictably,netizens quickly gather on the social platform dominated by microblog to participate and discuss with high enthusiasm.They become the digital projection of the real society.Through micro-blog publication,transmission and comment,the emergency has spread quickly,causing public opinion to break out and form emergency net public opinion.The negative emotions of netizens then appear in the phenomenon of group gathering on social platforms.In order to avoid the serious harm to society caused by the growth of negative public opinion,it is urgent to effectively relieve the negative emotions of netizens in emergencies,and reasonably monitor and guide the spread of online public opinion.It is noted that opinion leaders in social networks play a key role in regulating the emotional orientation of Internet users.Therefore,the identification and analysis of opinion leaders has become an important breakthrough in monitoring and controlling the public opinion spread of emergencies.Based on this,this paper starts from the user influence and emotional polarity,maps the size of user communication ability and emotional polarity state to the network emotional influence,identifies opinion leaders by building the Network Sentiment Influence(NSIF)model,and discusses the mechanism of its effect on netizens’ emotion in the public opinion communication of emergencies.In addition,this paper also constructs an emotional map to visually present and analyze the process of emergency online public opinion transmission,so as to timely and effectively understand,intervene and guide the public opinion transmission of emergencies,prevent the growth of malignant public opinion,and maintain social stability.The main research contents are divided into the following parts:First,the algorithm of user emotional polarity is studied.From the perspective of emotional dictionary and deep learning,four algorithms are constructed to identify emotional polarity,namely,the algorithm based on the extended polar dictionary,the optimized text convolutional neural networks(Text-CNN)、the optimized text recurrent neural network(Text-RNN)and BERT-Att-BiLSTM model integrating attention mechanism.In the method based on the extended dictionary,this paper expands the network word dictionary and the degree word dictionary on the basis of the existing dictionary,and designs the polarity calculation rules of polar phrases;In the optimization of Text-CNN,the full connection layer(Re LU as the activation function)is added in this paper to facilitate the extraction of richer features and prevent adverse effects caused by drastic dimension reduction;In the optimized Text-RNN,2-Max Pooling operation is adopted for all neuron outputs in this paper,so that the impact of features at other times on model classification can be considered at the same time;In the BERT-Att-BiLSTM model,The BERT model is used as a word vector model to express the semantic meaning of a word.And his paper introduces a two-way BiLSTM model,which integrates the attention mechanism and focuses on the key characteristics to increase the precision of classification.Second,construct the NSIF model to lay a theoretical foundation for calculating users’ emotional influence and identifying opinion leaders.This paper is based on the traditional spatial autoregressive model,drawing on the network influence model of Zou et al.(2021),and introducing exogenous independent variables to parameterize the autoregressive coefficients.Zou et al,the connection function is allowed to be unknown.A four-stage estimation method based on spline estimation and nonparametric two-stage least squares estimation(NP2SLS)is proposed to estimate the parameters and nonparametric parts of the model.The rationality and feasibility of this model are proved by numerical simulation.Thirdly,on the basis of NSIF model,this paper researches the application of public opinion dissemination and opinion leader recognition.In this paper,the "Rainstorm Event of July 20,Zhengzhou" is selected as a typical case.Firstly,combining the communication characteristics of microblog and user behavior data,the microblog opinion leader and user emotional orientation indicator system are constructed,and the microblog data related to microblog public opinion topics of emergencies are obtained and preprocessed using the crawler technology;Through comparative analysis,the accuracy of four emotion classification models is 70.21%,85.30%,86.45%,89.70%respectively,and BERT-Att-BiLSTM model is the optimal model.Then,according to the classification results of the optimal model,this paper proposes a quantitative method of emotion polarity value to obtain the user’s emotion polarity value;The polarity value of user’s emotion is taken as the response variable,and the user’s emotional influence is estimated using the constructed NSIF model;Identify opinion leaders based on the estimated emotional influence value,and verify the rationality of this model and its good applicability in emergencies by comparing it with traditional methods;Finally,this paper divides the life cycle of the emergency into "generation period,outbreak and diffusion period,and regression period",and constructs the emotional map of social network public opinion communication in different periods,thus effectively showing the emotional transmission mode and evolution characteristics of the emergency.