Research On Event Trigger Word Extraction Based On Convolutional Bidirectional Gated Recurrent Unit

In the face of the explosive growth of electronic texts,what people care about is how to intelligently process these data and information,and obtain the issues that users really care about from these texts.In this context,the concept of information extraction was proposed.Event extraction is a sub-direction of the field of information extraction,and it is also a research focus and difficulty in this field.Traditional event extraction methods based on pattern matching need to construct patterns manually,which consumes a lot of resources.Feature-based methods will rely excessively on natural language processing tools in the process of extracting features.This process will also consume a lot of human resources,and it is also prone to a series of problems such as data sparsity and error propagation.Nowadays,deep learning technology is popular in all walks of life,especially in the application of speech recognition and image processing.Affected by this,researchers have also begun to introduce deep learning technology in the field of NLP.The introduction of deep learning methods can reduce the dependence on natural language processing tools.This method can make full use of text context information to automatically mine effective features from text.People gradually began to apply deep learning technology to the field of event extraction.Considering the shortcomings of previous research models,we propose a new model to perform the task of trigger word extraction.The main idea of the model is to integrate the convolutional neural network and the bidirectional gated recurrent unit to better determine the event trigger word and the type of event.We first use Google's open source toolkit Word2 Vec to convert text data into word vectors so that the computer can better understand;then use convolutional neural networks to extract local word-level features of the input data,and then use the bidirectional gated recurrent unit Sentence-level feature extraction is performed on the input data.The bidirectional gated recurrent unit can combine the word dependence in two directions to better mine the semantic information of the sentence.Next,we will combine the extracted features of the two parts.In this way,the features we extract are richer,including both word-level features and sentence-level features of the text context;finally,the extracted features are put into the softmax layer for trigger word recognition judgment.By conducting experiments on the ACE2005 English corpus and the CEC Chinese Emergency Corpus,we compare the experimental results of the model proposed in this paper with those of the previous models and find that whether it is in terms of trigger word extraction effect or convergence speed,The models we put forward show better results.