Risk Analysis Of Online Public Opinion Of Emergency Based On Dynamic Bayesian Network

With the rapid development of mobile Internet,a bulk amount of online public opinions,such as public emotions,attitudes and opinions related to the event generated and then quickly and widely disseminated through social media in the aftermath of emergency events,and the orientation of online public opinions is closely related to the harmony and stability of the society.Social media not only provides real-time,convenient and effective information for disaster response during emergency situations,but also provides a fertile ground for extreme emotions,rumors and malicious misinformation simultaneously,thus resulting in various risks of online public opinion.In the era of mobile Internet,enhancing the risk management of online public opinion in emergency context has become an important planning and important practical demand in China at the present stage.This paper takes disaster system theory as the basic theoretical research framework,we obtain the post data related to emergency events and corresponding user data through Sina Weibo API,and aims to build a dynamic Bayesian network model for risk analysis through a combination of knowledge driven and data driven approaches.First of all,this paper introduces the concept of vulnerability in disaster research into online public opinion research,and propose the framework of social media users' vulnerability analysis.A lexicon-based approach is used to quantify the textual content of each post generated by user in numerical vectors,including emotional intensity and emotional word frequency count.Then,principal component analysis and ordered logistic regression analysis are employed to investigate what and how driving factors associated with users' vulnerability to online public opinion of emergency in social media.Secondly,the conceptual framework of online public opinion's risk in emergency is proposed from the perspective of disaster system theory,based on TF-IDF technology to identify key words related to disaster-causing factors,and then Seemingly Unrelated Regression(SUR)analysis is employed to quantitatively analyze the interplay rules between the risk factors.Finally,the initial Bayesian network model for the risk analysis of online public opinion in emergency is constructed based on the interplay rules,K2 structural learning algorithm and the EM parameter learning algorithm are employed to learn the final dynamic Bayesian network model from observation data by a data-drivenapproach,ten-fold cross-validation method is used to validate the performance of proposed model.The risk analysis model of online public opinion in emergency based on the dynamic Bayesian network proposed in this paper fully considers both the interplay rules between the risk factors of online public opinion and their time-dependent relationship,and can update the probability of each node in the network through the time sequence data.Compared with the traditional static Bayesian network model,the dynamic Bayesian network model shows a higher prediction accuracy,which provides a decision-making basis for the emergency management department of online public opinion to improve the pre-disposal time of online public opinion risk.