Research On The Method Of Emergency Event Extraction For Microblog

The occurrence of a sudden event will not only disrupt the social order,It has produced many problem seriously threatened our social stability.At the same time,it poses a severe test for public security to maintain social stability and protect the safety of people's property.With the rapid development of social media,The amount of emergency information on the Internet has increased dramatically.How to accurately identify and extract emergency information from a large amount of unstructured event information and analyze the public opinion trend of the emergency has always been highly valued by the public security department.This paper analyzes the existing emergency extraction methods in depth,and studies the two main aspects of emergency extraction,Event extraction and event type classification prediction in the Weibo,and proposes multi-classification.Incident extraction model and data weighted optimization of event classification methods have an active auxiliary role in extracting emergency information and reducing the workload of police.The main research work of the full text is as follows:Firstly,a new emergency extraction model is proposed based on the current situation of emergency extraction,and the relevant information of the emergency is extracted accordingly.Existing emergency extraction models mostly use cascading methods.The disadvantage is that the event information is insufficiently used,which is prone to cascading errors.At the same time,the grammatical and syntactic analysis is limited by unregistered words and expert experience.In this paper,useing mining part-of-speech features and semantic relationships in the text,and combining Google's new word vector training model-the BERT model,the multi-layer long-term short-term memory network LSTM structure is used to extract candidate event sentence features to generate candidate phrase vectors.The multi-classification method is used to replace the two-classification method to calculate and score the phrase vector.The event trigger word is determined according to the word score to implement event extraction.Multiple experiments on the Chinese emergency corpus and testing on the Weibo data set show that the model has a certain Effectiveness.Secondly,a density-based clustering method is used to extract the elements of the identified event content and classify and predict the types of emergencies.In order to ensure extraction efficiency,relevant event elements are extracted based on rule-matching methods,and then data cleaning is performed.The default event elements are filled and filled to form an emergency event data set,which is mathematically analyzed and filtered.Data and outliers to ensure the validity of the data.The data is normalized to reduce the dimensionality of the data to generate feature data,and then weighted and optimized to perform cluster analysis and encapsulate it intoa machine learning algorithm model to achieve emergencies.Type classification prediction,and several experiments on the Weibo dataset prove that the classification model has a good effect.Finally,comparative analysis and summary are made on the extraction of emergency events,the extraction of elements,and the prediction of event classification.The experiments were performed on Chinese emergency corpus and Weibo data set,and the experimental analysis,comparison analysis and induction were summarized.The methods proposed and adopted in the paper showed good results in the experimental results,which proved the effectiveness of the experimental methods in the article.