Weibo-based Event Extraction And Risk Assessment For Public Safety

As an online social media platform,Weibo developed fast.It owns a huge number of users and real-time public opinion information.Weibo has become a useful tool for researchers to discover public security-related information.Weibo-based public safety event extraction and risk assessment can quickly identify public safety events so that relevant staff could know the updated incidents.On the other hand,administrators could understand the risk status of different public safety events through risk assessment so as to effectively carry out pre-arranged mitigation plan.In this thesis,event extraction algorithm is adopted to Chinese texts in the field of public safety.The aim of this research is to find Weibo-suitable text extraction method and apply it to the field of public safety.We adopt algorithm base on burst words to extraction public safety events after compared it with the document-based algorithm.This paper analyses the difference between public safety and non-public safety events.For example,public safety event extraction requires higher real-time performance and decision-makers tend to focus on only events that happen in a certain area under administration,so the required information are related to location.In the method based on the burst words,different Weibo are classified according to the province.The burst words were filtered out according to frequency difference and expressed by vectors.Then,Single-pass clustering algorithm is used to cluster the burst words into different categories.Finally,similar clusters are merged to obtain more accurate event clusters.The choice of algorithm takes the real-time property of public safety events into account.Besides,using the public dataset from Fudan University,this thesis compares the results of the burst words-based event extraction method with the event extraction method based on documents.The experiment shows that the accuracy and recall rate of the burst words-based algorithm are higher than that of the document-based method and the calculation efficiency is higher.After safety events extraction,this paper proposes an public safety risk assessment model based on influence and sentiment.For Weibo data,influence and sentiment are adopted to be two features,which can not only be directly obtained or calculated through Weibo data,but also reflect the development trend of an event and people's attitude toward it so as to evaluate the risk.In this thesis,using the events data from the extraction result and COVID-19 Weibo data,hybrid method including matrix grading method and fuzzycomprehensive evaluation method are used to assess the risks of public safety events.Page Rank is adopted with user behaviour index to measure influence.When computing sentiment,the Valence-Arousal binary emotion space representation is applied and the regression model is based on SVM and LSTM.Then the emotional polarity values and strength value are obtained.Finally,after obtaining the relevant Weibo posts after the extraction of crisis events,the qualitative and quantitative evaluation algorithms were applied and compared.The risk levels of different events were obtained as the model result.The result shows that the two methods are in line with the reality,and the matrix grading method is more intuitive.Also,the result demonstrates the effectiveness of the proposed indicators-influence and sentiment.Finally,a public safety events extraction and risk assessment system is designed and tested.The feasibility of the event extraction and risk assessment algorithm is verified.