Research On Named Entity Recognition And Entity Relationship Extraction Of Medical Data Text Based On Attention

Natural Language Processing purpose is to allow machines to understand natural language text processing information instead of human beings,reduce labor costs,and information extraction is one of the hot spots,deep learning application in information extraction result obtained is all the technology in the best.However,there are still shortcomings as a research on information extraction in deep learning.Extraction for named entity recognition extracts information entity relationships with the core information extraction will directly affect the performance of the machine to extract the key information for the text,thus affecting subsequent task is determined,these two tasks is the key to all of the natural language processing application.Therefore,this paper studies the named entity recognition model and entity relationship extraction model based on the clinical medical field.Aiming at the problem of incomplete expression of contextual relationship between word vector representation in existing deep learning models,this paper establishes a named entity recognition model in the task of named entity recognition in the clinical medical field.The named entity recognition research discussed in this article based on the improvement of the BioBERT model.It mainly uses the last three layers of feature vectors generated in the BioBERT training process to complete LSTM processing,and then builds the relationship between BioBERT and LSTM through residual links,and finally completes the dependency through CRF Sexual treatment gets the result.This model achieves good results in the BC2GM-Disease data set and the BC5CDR-Disease dataset.BERT excellent model for the current use of existing hardware issues such harsh conditions,establish a solid relationship extraction model in the field of clinical entity relation extraction task.Recognition entity relationship is discussed herein primarily in improved coding module.The model is first to enter text into the word vector representation,generating a vector machine can be appreciated,as used herein,is a pretrained model Glove.The research uses the attention mechanism-based Sparse Transformer,Bil STM,Capsule,and K-Maxpooling to further screen features.Finally,the obtained features are transmitted to the Softmax prediction answer,and the prepared answer with the highest probability is found as the predicted answer,and the optimal effect is achieved in the DDI dataset and CPI dataset.