Research On Emergency Organization Identification And Relationship Extraction For Emergency Response Plans

Emergencies occurring in modern society usually have the characteristics of large scale and high complexity,which government needs to set up reasonable institutions and take corresponding specific measures to deal with emergencies.Therefore,each province has developed a variety of types and accurate emergency plan system.To analyze the text of an unconventional emergency response plans,the premise of using the information of emergency disposal quickly is to complete the extraction of the relevant organizations and the relationship between them.At present,research methods of named entity recognition and relationship extraction mainly include rule-based,statistical machine learning and deep learning methods.The recognition research in many fields has been mature and achieved good results,but the research on Chinese emergency plan text is less.This paper studies the emergency organization identification and relationship extraction of Chinese emergency plan.The main work of this paper are as follows:(1)We crawl the emergency response plans texts of various provincial people's government websites,and build a large-scale emergency response plans data set,which lays the foundation for the next work.(2)In the task of emergency organizations identification,firstlly we learn the existing method based on Conditional Random Field(CRF)to identify emergency organizations and found that the method based on CRF cannot accurately ues the surrounding information of organization entity,which the effect of identification on the complex emergency organization names is not ideal.Based on this we designed BiGRU-Att-CRF model to identify emergency organizations.Through the comparison of experimental results,in the task of emergency organization identification,the F1 value of the method based on B,M and O labels is reached 86%,86.9%,93.5%respectively.(3)In the task of emergency organization relationship extraction,this paper introduces the Bidirectional Encoder Representation from Transformers(BERT)to design the Bidirectional Long Short Term Memory Networks based on attention mechanism,BiLSTM)emergency organization relationship extraction method(BERT-BiLSTM-Att),which uses the pre training language model of Bert in the word vector training stage,improves the representation ability of word vector,and uses attention mechanism to assign reasonable attention weight to the model.Compared with other models,this method has the best performance in semantic information extraction,and the recall rate and F1 value in emergency organization relationship extraction are 75.6% and 73.8% respectively.