Research On Drug-drug Interaction Extraction Based On Deep Learning

In recent years,technologies such as big data and artificial intelligence have developed rapidly.Processing massive text data quickly and efficiently through natural language processing has become an important means to deal with the explosive growth of data in the era of big data.The rapid development of biomedicine also promotes the rapid growth of biomedical literature.Large number of unstructured or semi-structured biomedical literature contains rich and valuable biomedical knowledge.It is of great significance to extract knowledge from biomedical texts efficiently and automatically.In biomedical research,include many biomedical named entities,such as proteins,diseases,drugs,DNA and RNA.And many relationships between these entities,such as the interaction between proteins and drugs.Drug interaction relationship extraction aims to extract the relationship types between two drug entities in biomedical texts.The study of drug interaction relationship is of great significance for guiding clinical medication and ensuring the safety of patients.Deep learning method is the main stream method applied to extraction drug interaction relationship,such as convolutional neural network,recurrent neural network and its variants.Convolutional neural network and recurrent neural network have the problem of insufficient feature extraction when used alone due to the characteristics of their own structure of the model.To solve this problem,this dissertation proposes a Bi GRU and CNN fusion bilayer model,Bi GRU layer extracts context dependent features of text,CNN layer extracts local details of text.The input vector matrix of the models is constructed by word2 vec tool.After experimental evaluation,the fusion model based on Bi GRU and CNN achieves a comprehensive evaluation rate of 75%.Compared with the single-layer neural network model,the performance of the model is greatly improved,which can effectively extract the action relationship of drug entities.In the distributed representation of text,the feature vectors represented by word vectors only stay in the shallow part of speech features such as words,vocabulary and case,and the dependency syntax of text sentences is not effectively used.To solve this problem,a dual channel convolutional neural network is proposed,and dependency information and drug description information are added as auxiliary drug knowledge database.In the dual channel convolution network,the input of the model is the word vector plus drug description and the word vector plus dependency information.By setting up a control experiment,explore the improvement ability of dependency information and drug description information on the extraction performance of the model.The experimental results show that the comprehensive evaluation rate of the drug interaction relationship extraction model based on dual channel convolution fusion of dependency information and drug description information is 80.59%,and the dependency information and drug description information can effectively improve the extraction performance.The method of extracting drug-drug interaction relationship proposed in this dissertation has excellent performance in entity relationship recognition.It can automatically and efficiently extract drug entity relationship pairs from vast unstructured or semi-structured biomedical literature,thus laying a solid foundation for guiding clinical medication and ensuring drug safety.