Research On Algorithm For The Evolution Of Emergency Situation In Omin-Media Environment

In recent years,our country has experienced frequent social security incidents,posing a significant threat to national security and social stability.In the context of omnimedia,the dissemination of public opinion information on emergencies presents the characteristics of diverse subjects,varied forms,and rapid diffusion.Under the influence of public opinion,the situation of social security incidents may worsen,leading to more severe consequences.However,the current research on the evolution of emergencies has limitations,such as one-sided analysis of influencing factors,ignoring the interference of data characteristics on prediction models,a lack of quantitative research on the risk of violent terrorism,and the interpretability of results.Therefore,this thesis takes social security emergencies as the research object and constructs a model around the two aspects of public opinion trend prediction and the risk of violent terrorism prediction.The main research results and innovations are as follows:(1)To address the issue of one-sided analysis of the influencing factors in the evolution of emergencies,the process of emergency situation evolution is attributed to five components in the "5W" model: participants,event incentives,content manifestation,time features and space features.According to the analysis of the evolution of public opinion and the process of violent terrorist incidents,the five components are subdivided into 12 and 17 influencing feature factors respectively,to establish systems of features for the evolution of public opinion and the risk of violent terrorism as support for the data index in the prediction model.(2)For the time series data of public opinion,it has the characteristics of nonstationarity and correlation between features.Therefore,this thesis uses the complete ensemble empirical mode decomposition with adaptive noise method to decompose the non-stationary public opinion data into a combination of multiple locally stationary sequences.Then,the method of factor analysis is employed to identify the hidden structure and pattern of the data in order to explore the correlation among multiple factors.Finally,an attention mechanism is introduced into the temporal convolutional neural network to predict the trend of public opinion.Comparison with multiple models shows that the public opinion trend prediction model based on multi-dimensional nonstationarity factors proposed in this thesis has significant advantages in performance indicators,such as RMSE,MAE,and R2.(3)Considering the lack of quantitative research and interpretability of the risk of violent terrorism in emergencies,this thesis proposes a combined weighting method based on term frequency-inverse document frequency and triangular fuzzy numbers for calculating the weight of feature factors,and uses bayesian network to carry out causal inference,thus predicting the risk probability of violent terrorism in the evolution of emergency situation.The key factors that trigger the risk are analyzed through reverse inference,and verify the reliability and applicability of the model in multiple real cases.