Research On Drug Interaction Extraction Based On BiLSTM-CRF With Attention Mechanism

Drug-Drug interaction is happened when patient take two drugs together,one drug could influence absorption,metabolism of another drug.It could enhance or weaken the effect of drugs,sometimes it could produce side effects.These side effects are very dangerous,sometimes it could cause life danger.Database like DrugBank and Micromedex are summarized by professionals,they contain some DDIs,but most of DDIs are still hidden in medical literature.Therefore,how to extract DDI rapidly and efficiently from medical literature is one of the emphases for research.In early stage,the extraction of DDI was based on hand-crafted rules,this method was effective for simply expressed statements,but not for complicated statements.Later,feature-based and kernel-based methods were used to extract DDI.Deep learning,one of the latest trends of machine learning and artificial intelligence,has brought revolutionary advance in many fields.A growing number of researchers using deep learning techniques to extract DDI.Based on deep learning methods,filtering samples was used to overcome the imbalance corpus problem.LSTM was used to overcome the long sentences problem in corpus.Besides,CRF was used as classifier to get final results because I considered context dependencies in corpus.To improve the efficiency of DDI extraction,I propose a BiLSTM-CRF based on attention mechanism.The corpus I used in this research is corpus of the task called Extraction of DrugDrug Interactions from BioMedical Texts in SemEval 2013.Due to positive samples and negative samples in corpus which differed widely in scale.In order to find out the impact of whether filtering negative samples,I conducted comparison experiment for screened negative samples and unscreened negative samples.It shows after preprocessing of screened negative samples,it enhances the efficiency of the model.Besides,in order to classify the efficiency of the proposed model,comparison experiments in different types of Attention mechanism,different types of classifiers and different types of models were conducted.