Research On Military Event Extraction Based On Deep Learning

The advancement of science and technology has promoted the rapid change of military weaponry and equipment and the maturing of automated information warfare operations has marked the official arrival of the era of highly informative military in our country.During the daily military training and operations,a great deal of operational information has been formed.How to automate the extraction of structural information of physical events in texts is a difficult and significant issue.The former method is based on statistics machine learning method.The statistics-based machine learning method is more mature,but the features based can not well reflect the general law of language semantics,so the effect is poor.The method based on deep learning has made a breakthrough in semantics,and it has been tried and achieved good results in the model and effect.The multi-dimensional feature representation proposed in this paper and the military equipment entity event extraction model of hierarchical depth network model effectively improve the semantic representation ability of word representation on a variety of military information text mining tasks.In this paper,a two-stage extraction model is proposed.Firstly,event-triggered words are identified by using instance representation and recurrent neural networks.Event elements are identified by using models of first-stage output and local features and short-term memory networks.The experiment found that the fusion of word vectors,syntactic analysis results,part of speech,entity eigenvectors,weapon equipment naming entities and other characteristics,to provide high quality input for the text mining system in the field of military information,which can effectively enhance the overall effect of event extraction.In summary,this paper semantically and semantically from the perspective of semantic and multidimensional features,training the semantic generation model for the field of military information and using the word vector for the revolving neural network based military action event trigger word extraction model and the dynamic event element extraction model,Achieved a good relationship between the extraction effect,in the future work will be based on the military information in the event extraction model to further improve.