Research On Event Detection And Prediction Based On Deep Learning

With the rapid development of information technology,massive text information appears every day.It is time-consuming and laborious to mine the rich knowledge contained in the information manually.How to extract and use the knowledge from the massive information quickly and efficiently has become a hot topic.As a technology of extracting structural information from unstructured text,information extraction arises at the historic moment.Event extraction is a very important task in information extraction,and it is attracting more and more researchers’ attention.Event extraction includes two tasks: event detection and argument extraction.Firstly,the main task of event detection is to classify events based on trigger words in text.At present,event detection based on deep learning method relies heavily on a large number of training corpus.The event detection is defined into few shot learning,and the prototypical network is used to detect the event under the background of sparse annotation.The most representative information of event type is trigger words.The previous method of fusing the relative position information of trigger words makes the feature information of trigger words not fully utilized.For this reason,the feature of trigger words is processed separately,and a loss function based on trigger words is added.This paper proposes an event detection model based on hybrid Attention-Based prototypical networks with independent trigger information(HATTPNT),which improves the robustness of the event detection model under few shot learning.Secondly,the event chain extracted from unstructured text is regarded as a known event in the event prediction task,and the subsequent events are predicted in the form of multiple-choice questions.Compared with the structure of event pair or event chain,this paper introduces more complex graph structure information between events to enrich the relationship semantics between events,and uses gated graph neural network to extract features of structural events.At the same time,considering that in the subsequent event step of reasoning,the attention mechanism based on convolutional neural network(CNN)is used to learn the weight of known events to improve the prediction accuracy of event prediction.This paper presents an event prediction model based on gate graph neural network and CNN Attention-Based(GGNN-CATT).Finally,the proposed HATTPNT model and GGNN-CATT model respectively compared and analyzed with other methods on different public datasets,and the effectiveness and the advanced nature of the two models are verified.