Application Of Deep Neural Network In Biomedical Information Extraction

With the rapid development of the Internet and advances in biomedical research,the number of biomedical documents has exploded in recent years.At the same time,the development of social media has also provided a wealth of valuable resources for the biomedical field.Most of these biomedical texts are stored as an unstructured form,these unstructured data contain a large amount of valuable knowledge.Among them,drug entities often serve as the main research object for scholars in the field of biomedical research.The emergence of drugs has brought good news to people's life and health,but the adverse reactions caused by drugs are also extremely dangerous.The extraction of structured information about drug entities in the massive unstructured biomedical literature has been a hot research issue for researchers.At the same time,the development of deep learning methods in recent years has provided new opportunities for the study of drug information extraction.If this research is done,it can expand the existing drug knowledge base,and it can also facilitate related personnel to obtain the desired knowledge from a large number of documents more quickly.This will play an important role for researchers and medical practitioners in related fields.Therefore,information extraction of drugs in the biomedical field is currently the hottest and most important research topic in the field.The information extraction of drugs in this article mainly focuses on three aspects of research:classification of adverse drug reactions,drug name entities recognition and drug relationship extraction.The research content mainly includes the following parts:The classification of adverse drug reaction based on hybrid neural network model with multi-feature.We use the part of speech features,sentiment features and position features to construct vectors as inputs for neural network models,combined with different features to solve the problem of limited characterization information for single features.Through the neural network models,the classification task of adverse drug reactions was completed.Drug name entity recognition method based on CNN+LSTM+CRF.Using character-based vectors and word-based vectors as inputs,we construct a learning model.This model uses the deep learning method to extract features and avoids the artificial construction of complex features.Then,it uses the CRF method,the best performance known in drug name entity recognition as the prediction method to improve the model performance.Drug relationship extraction based on CNN+LSTM.This paper does not construct a large number of artificial features,only use the position features to build feature vectors,we use CNN and LSTM to build a neural network model.Experiments show that the proposed model has better performance and better results.Finally,this paper also uses the knowledge map to build a visualization system to intuitively demonstrate the results of drug relationship extraction.