Biomedical Event Extraction Mechanism Based On Combined Deep Learning

Biomedical event extraction is the focus of biomedical information extraction.It studies the relationship and change of proteins at the molecular level,and finds out their types and participating elements.At present,many common event extraction mechanisms have achieved good results in the extraction of biomedical events,but compared with other fields,the precision rates and recall rates are generally low,there is a lot of room for improvement.This paper proposes a mechanism for extracting biomedical events using a combined deep learning model.It makes full use of the advantages of different models,extracts features in both timing and spatial structure of corpus,and achieves high precision rates.Follow the classic pipelined extraction process: pretreatment,feature extraction,trigger word recognition,event element detection,rule post-processing.Firstly,according to the definition of biomedical event,the original corpus is preprocessed to make the corpus more suitable for decimation task;Using the distributed semantic word vector,combined with the basic dependency structural features and location features,the statement expression vector is generated.Then the recurrent neural network model is used to extract the temporal characteristics of the sentence;Using a convolutional neural network model to classify,identify trigger words and event categories;In the event element detection phase,simple events and complex events are extracted at the same time.Based on the deep features and triggers obtained in the previous step,the convolutional neural network is used to identify the event elements.Identify all relationship pairs at once,and generate simple events and complex events based on the categories of relationships;Finally,combining the field characteristics of biomedical events,define rules to further optimize the results of element detection and generate structured biomedical events.This article experiments on the corpus provided by the BioNLP'09 shared task,and achieves high accuracy.The experimental results show that the proposed method is feasible in biomedical event extraction tasks.