Research And Implementation Of Biomedical Event Extraction Method

With the rapid spread of the Internet,the number of biomedical literature is increasing quickly,which has made it difficult to obtain useful knowledge from massive unstructured biomedical literature for researchers to use.Therefore,biomedical information extraction technology came into being.Biomedical event extraction is an important and basic task in biomedical information extraction,which aims to extract the multiple semantic relationships between fine-grained biological entities and show the events in a structured form.This task is of great significance for drug development and disease prevention.So,based on MLEE corpus,we use methods based on neural network to carry out experimental research on biomedical event extraction.The main research works are as follows:(1)A method based on Graph Convolutional Network is proposed for biomedical event trigger identification.We perform experiments based on MLEE corpus,and obtain Part-Of-Speech and dependency syntactic information through natural language processing tools to enrich the initial feature representation.We use Bidirectional Long Short Term Memory Network(Bi LSTM)to obtain context representation and Graph Convolutional Network(GCN)to model dependent information.A softmax function is for the classification of trigger.So far,the task of biomedical event trigger identification is completed.The experimental results verify the effectiveness of the proposed method.(2)A method based on Combined Neural Network is proposed for biomedical event argument detection.According to the definition of biomedical events,we construct candidate event trigger-argument pairs and use Convolutional Neural Network(CNN)to extract local features with convolution operations to better learn representations.A Gated Recurrent Neural Network(GRU)is used to extract global context features and for further feature extraction.Attention mechanism is also introduced to fuse local and global information.We classify candidate pairs using softmax function,and complete argument detection.Finally,the final event is obtained by post-processing with some hand-crafted features.Experimental results show that the proposed method can get good performance.(3)A method based on Graph Attention Network is proposed for joint extraction of biomedical events.The word vectors are obtained through language model training,which is combined with the Part-Of-Speech vector,entity type vector,and position vector to obtain the input representation.We use Bi LSTM to extract context features and enrich word representations.The corpus is preprocessed to obtain some information such as dependency parsing and the syntactic structure obtained from the dependency parsing is used to extract deep structure features with Graph Attention Network(GAT).Finally,biomedical event trigger identification and biomedical event argument detection are performed through the softmax classifier to complete the joint extraction of biomedical events.The experimental results show that the joint model,to some extent,can alleviate the cascade error transmission and can obtain better biomedical event extraction performance.