Rearch On English Event Extraction Based On Deep Learning

Event is an objective fact that occurs at a particular time and in the environment and is represented by a number of roles.Event extraction requires automatic identification and extraction of structured information containing event types,event elements,and event roles from unstructured source text containing event information.The traditional method of extracting the extract is highly dependent on the characteristics of artificial selection.On the one hand,the artifacts need to extract the characteristics of the experts,and on the other hand,the system adaptability is poor and the generalization ability is low.In recent years,researchers began to use the depth of learning methods to automatically learn the characteristics of this subject for how to apply the deep learning to event extraction task deeply,the main research includes the following three aspects:Optimizing Event Extraction with Deep Semantic FeaturesAt present,many methods use the shallow lexical level feature,it is difficult to obtain the text of the deeper semantic information,training model of the generalization of the weak.On the basis of the traditional shallow learning characteristics,this topic introduces the deep semantic features and improves the generalization ability of the model.The experimental results show that the addition of deep semantic features improves the event extraction performance of the model.Research on Event Extraction Based on Neural NetworkThe traditional feature-based event extraction method relies on artificially designed features and complex natural language processing tools.Although these methods proved to be effective by experiment,but spent a lot of artificial energy and there will be sparse data problems.This topic uses the neural network based event extraction method,can not use a large number of artificial design features and automatically learn the characteristics of the sentence.In addition,this paper proposes an event extraction model of combined convolution neural network and bidirectional cyclic neural network.Combining the advantages of convolution neural network and cyclic neural network,event trigger word recognition and event type classification are proposed.Combining Context Dependency and Sentence Semantic for Event DetectionThe existing method ignores the dependency of the word to be measured and the semantic information of the sentence,which is helpful for the detection of event clues.In this paper,a neural network method is proposed,which uses the bidirectional length-and-time memory network to capture the context dependency of the word to be tested in the sentence.At the same time,the semantic representation of the sentence is studied by using the gated cycle neural network,and the two kinds of information are improved.Compared with the existing neural network method,this method can effectively use the context-dependent and semantic information of sentences to improve the performance of event clue detection.The method can not use any natural language processing tool to avoid error transmission.