Biomedical Entity And Relation Extraction Based On Deep Learning

With the continuous development of the biomedical field,a large number of academic and scientific research achievements are displayed in the form of papers and documents.Therefore,a vast amount of biomedical literature has formed a sea of knowledge,which needs to be explored and studied by scholars.Biomedical Entity and Relation Extraction is to extract biomedical entities and the relations between entities automatically from biomedical literature.It can help researchers discover and learn relevant knowledge efficiently,to further promote the development of the biomedical field.For biomedical named entity recognition,this paper proposes a method based on reinforcement training called “REIN-NER”.REIN-NER defines the named entity recognition task as a word-level classification problem.Firstly,the basic recognition model(P-net)is pre-trained to obtain the ability to “capture the dependency between input and output”.Then the “Scoring System” is designed to give the model feedback,and reinforcement training is carried out to make the model further obtain the ability to “capture the dependency between the output tag sequences”.Compared with these current popular methods,REIN-NER does not use the pre-training model with a large number of parameters or the CRF algorithm.It only uses Bi-LSTM and reinforcement training mechanism to make the model have excellent performance in two disease recognition tasks and a chemical recognition task.For Biomedical Relation Extraction,this paper proposes a method based on the triplet-loss training strategy called “TL-BERT”.TL-BERT first uses the designed triple-data generation rules to obtain triple-data.Then the sentence-level and entity-level features are extracted from the triple-data.Finally,based on these features,the triple-loss and cross-entropy loss are calculated to train the model.Compared with other relation extraction methods,TL-BERT creatively uses triplet-loss to train the model,it solves the problem of“instances from the same sentence but belonging to different classes are difficult to classify”and improves the performance of relation extraction.This method has achieved excellent performance in two protein-protein interaction extraction tasks and a drug-drug interaction extraction task.