Extracting Biomedical Events With Parallel Multi-pooling Convolutional Neural Network

A vast and ever-expanding body of natural language text is becoming increasingly difficult to leverage.This is particularly true in the domain of biomedical research articles which are increasing exponentially.Consequently,the need to extract interested and structured information automatically from biomedical text continues to grow.Event extraction using expressive structured representations has been a significant focus of recent efforts in biomedical information extraction.Generally,a biomedical event includes two main parts,the biomedical event trigger and its argument(s),where triggers are usually verbal forms or nominalizations of verbs and arguments are often biomedical entities or other events.Traditionally,most of the state-of-the-art systems have been based on shallow machine-learning methods which require many complex hand-designed features.In addition,the words encoded by one-hot are unable to represent the semantic information.Therefore,we utilize dependency-based embeddings to represent words semantically and syntactically.Then,a parallel multi-pooling convolutional neural network(PMCNN)model is proposed to capture the compositional semantic features of sentences,and the softmax classifier is utilized for classification.Furthermore,a rectified linear unit,which creates sparse representations with true zeros and is adapted to the biomedical event extraction,is employed as nonlinear functions in PMCNN architecture.Eventually,the SVM-based post-processing method is used to learn the correct combination of predicted argrumets to satisfy the biomedical events structure.The experimental results from MLEE dataset show that our approach achieves an F1-score of 80.27% in significant trigger classes and an F1-score of 59.65% in biomedical event extraction,which performs better than other state-of-the-art methods.