Research On Element Recognition For Chinese Emergencies

The world is in motion,and the world in motion is made up of events.Events can promote the change of things and the development of relations,and also constitute the basic unit of human understanding of the world.Emergency refers to the natural disasters,accidents,public health events and social security events that occur suddenly and cause serious social harm and need to be dealt with by emergency measures.In recent years,emergencies such as COVID-19 outbreak and Sino-US conflict are frequent.It has become particularly important to use event element recognition technology to extract emergencies from massive texts of emergencies.In the current research on Chinese emergency element recognition,most scholars often ignore the polysemy problem in the pre-training processing stage and the problem that the semantic of word vector features cannot fit the context.In this paper,by analyzing the vector features in emergency text,deep learning technology combined with event semantic information is used to study the recognition of emergency elements.Aiming at the polysemy problem in the pre-training processing stage,this paper first proposes an event element recognition model based on ALBERT-Bi LSTM-Attention.Then,on the basis of this model,considering the problem that the semantic information of word vectors does not meet the context,Furthermore,an event element recognition model of IDCNN-Bi LSTM-CBAM with rich semantic dimensions is proposed.The research content of this paper mainly includes the following two aspects:(1)Aiming at the polysemy problem in the pre-training processing stage,the event element recognition research based on ALBERT-Bi LSTM-Attention model is proposed.Firstly,in the pre-training stage,the Word2 vec model is introduced to train and reconstruct the emergent text,and the word vector representing the word and the relationship between the words is generated.Then the word vector is encoded at the character level with the ALBERT pre-training language model and fused into the joint word vector to solve the polysemous problem.Then,the joint word vector is input into the joint neural network combining Bi LSTM(Bi-directional Long Short-term Memory)and attention mechanism.Firstly,the contextual semantic feature information of the emergency text is extracted through Bi LSTM layer.Then,feature weights with different importance levels were allocated through the attention mechanism layer.Finally,softmax logistic regression layer is used for classification and processing to complete the task of emergency element identification.The model is tested on CEC2.0corpus and the F1 value of 84.33% is obtained,which proves that the model has good recognition effect.(2)Aiming at the problem that the semantic information of word vector does not conform to the context,the event element recognition research of IDCNN-Bi GRU-CBAM model with rich semantic dimensions is proposed.On the basis of model(1),considering the problem that the semantic information of word vectors does not conform to the context,a recognition model of IDCNN-Bi GRU-CBAM with rich semantic dimensions is proposed.Firstly,the text features of emergencies are analyzed and the partof speech analysis,block analysis,dependency syntax analysis and semantic role analysis are carried out to obtain the partof speech features,block features,dependency syntax features and semantic role features.And connected with the joint word vector generated by model(1),so as to enrich the semantic information of the joint word vector,enhance the expression ability of various word features,and make them conform to the context;Secondly,IDCNN(Iterated Dilated CNN)and Bi GRU are combined into a joint neural network,which can better extract local semantic information and global semantic information.Then,the attention mechanism is integrated into the convolutional neural network to filter out the feature information that does not conform to the context and give more weight to the important semantic information.Finally,softmax classifier is used for classification,and various event elements are obtained.The model is tested on CEC2.0 corpus and the F1 value of 88.01% is obtained,which proves that the model has a good effect on the recognition of emergent event elements.Figure [21] Table [13] Reference [59]...