Research On Emergency Recognition Based On Deep Learning

At present,more and more redundant data accumulates in the Internet,and seriously affects people to get event information that they need.How to make good use of big data,and extract the information,so that we can dig out the hidden value behind a large amount of data.This is the research direction in the field of event recognition.Through traditional pattern recognition and machine learning methods to deal with emergency event recognition,the performance of the model is relatively limited when the model deals with complex problems.For the extraction of deep-level event information features,the generalization and feature extraction ability of model are restricted to certain extent.Aiming at the above problems,the thesis uses deep neural network as the main research method to carry out technical research on the recognition of emergency event types and recognition of emergency event elements.The thesis mainly completed the following tasks:(1)In terms of emergency event types recognition,aiming at the weak generalization of traditional event recognition methods,the limitation of dependence on special knowledge of field,the longer train time of deep recurrent neural network,and the problem of gradient dispersion,the neural network joint model,Conv-RDBi GRU,integrated residual structure was proposed.Firstly,text corpus is preprocessed by word segmentation and stop words operation,and the processed data is trained to generate word vectors matrix by using word embedding.The local semantic features are extracted through convolution operation,and then deep context semantic features are extracted through RDBi GRU.Finally,the learned features are activated by softmax function and the recognition results are output.The Conv-RDBi GRU model is used to conduct experiments on the CEC dataset and we-media dataset crawled from network,and the simulation results show that this method improves precision and recall of emergency event recognition,and the F-value is better than other methods used in comparative experiment.(2)In terms of emergency event elements recognition,in view of the poor interpretability of recurrent neural network in respect of information features with different degrees of importance.The emergency event elements recognition method based on Bi GRU-AM model with extended semantic dimension is proposed.Firstly,text corpus is preprocessed by word segmentation and stop words operation,and the processed data is trained to generate word vectors matrix by using word embedding.The word vectors connect semantic features about part of speech,dependent syntactic relations and et al.Contextual information features are extracted through Bi GRU network,and then attention mechanism is integrated into Bi GRU network to make feature extraction more selective.Finally,the learned features are activated by softmax function to output recognition results.The Bi GRU-AM model with extended semantic dimensions is used to perform experiments in the CEC dataset.Simulation results show that compared with other algorithm of shallow machine learning,the model can effectively deal with the task of emergency event elements recognition,and the better Fvalue can be achieved compared with other methods used in comparative experiment.