Analysis Of Public Opinion Heat Of Public Health Emergencies Based On CEEMDAN-ARIMA-LSTM Combined Model

With the development of the Internet and information technology,the identification and monitoring of network public opinion has become a research issue that has attracted much attention.The public opinion heat is an extremely important indicator that reflects the public’s attention to social events.It can effectively reflect the social influence brought about by the corresponding event.Studying the changing range and fluctuation trend of public opinion enthusiasm can help the government to grasp the trend of public opinion in a timely manner.Controlling the dissemination of malignant public opinions and formulating intervention policies such as public opinion guidance provide practical and feasible basis.The monitoring and trend prediction of public opinion heat in public health emergencies not only helps to detect the epidemic situation in time,but also has important value in guiding the public to correctly face the epidemic situation,identify rumors,and avoid panic.It is an important object of network public opinion monitoring.This paper takes the online public opinion heat of the new crown epidemic as an example to explore the prediction method of the network public opinion heat of public health emergencies.The public opinion heat can usually be measured by the search popularity provided by the search engine,and the search popularity is bound to be affected by many external factors,so that the public opinion heat has both linear and nonlinear properties.The previous single models such as ARIMA and neural network have obvious shortcomings in forecasting.For example,the ARIMA model will produce large errors when predicting nonlinear time series.This paper proposes the CEEMDAN-ARIMA-LSTM combination model.First,the public opinion heat is decomposed using CEEMDAN and the corresponding sequence features are extracted.Then,for the decomposed subsequences,the ARIMA model and the LSTM model are used to successively analyze the linear and nonlinear parts of the sequence.Prediction improves prediction accuracy.In the empirical part,this paper obtains the public opinion data of the new crown epidemic from December 31,2019 to September 30,2020 based on the Baidu index,and weights the public opinion data of each keyword based on the entropy method to obtain the new coronavirus.Public opinion heat sequence,and divide the public opinion life cycle based on the public opinion life cycle theory,and use ARIMA,LSTM,CEEMDAN-ARIMA,CEEMDAN-LSTM,CEEMDAN-ARIMA-LSTM for modeling in stages,and based on multiple index values The modeling effect of the model is evaluated.The model prediction effect evaluation based on multiple evaluation indicators shows that,compared with the single model,the CEEMDAN-ARIMALSTM model can judge the public opinion heat more accurately,which confirms to a certain extent that the hybrid model has certain reliability in the field of public opinion enthusiasm..This study enriches the existing public opinion heat prediction methods,and provides a reference for the government to understand the public opinion heat trend of public health emergencies in a timely manner.