Research And Application Of Chinese Event Extraction Technology Based On GCNN Model

In recent years,with the rapid improvement of Internet technology,more and more information has poured into the Internet.The messy information makes people begin to rely on information extraction technology,and Event extraction belongs to information extraction.Event extraction contains two core sub-tasks: event trigger extraction and event argument extraction.Among them,event trigger word extraction is to find out the core words that can best reflect the occurrence of the event and classify them,while event argument extraction is to extract the important information such as the time,place and people involved in the event.The task of extracting Chinese events is mainly focused on conflict events,financial events,military events and so on.With the upsurge of "smart party building",the event extraction of party building data has attracted wide attention.The research in this field has important value for the retrieval of party building news and the popularization of party building related knowledge.In this paper,an event extraction corpus for Chinese party building data is constructed,and an event trigger extraction model and an event argument extraction model are proposed based on GCNN.Firstly,the model uses pre-training for text vectorization,next extracts the feature vectors through the gated convolutional layer,and then gives higher weight to the key features through the self-attention mechanism,afterwards performs dynamic multi-pooling according to the positions of the candidate word and event trigger,and finally output the result of event extraction through the output layer,and according to the loss distribution of each batch of training data,change the static loss function to a dynamic loss function.Compared with other benchmark models,the model proposed in this paper has obtained the optimal results in the event extraction experiment of Chinese party building data,which shows the effectiveness of GCNN model and dynamic loss function.In addition,this paper stores the results of event extraction into the database,and provides an event query system,which can query the event information according to the event type and event argument.