Research On Named Entity Recognition Technology In Biomedical Field

Recently,named entity recognition has attracted much attention in biomedical field.It is difficult for experts in biomedical field to keep up with the growth of biomedical literature with the existing ways of obtaining literature information.Therefore,higher requirements are put forward for text mining methods.A fundamental and essential task of text mining is named entity recognition.Biomedical Named Entity Recognition aims to identify and label specific types of entities from biomedical texts,such as genes,diseases,chemicals,etc.This will provide preconditions for subsequent work such as information extraction and question and answer system.Compared with general named entity recognition,biomedical named entity is often more challenging and potential research space.The research work of biomedical named entity recognition in this paper mainly includes the following two parts:(1)Biomedical named entity recognition based on contextualized capsule network.There are a large number of rare and unseen entities in the biomedical training texts because of sparse data.Lack of enough training instances and scarcity of information about rare and unseen entities limit the further improvement performance of Biomedical named entity recognition.In order to solve this problem,this paper proposes a biomedical named entity recognition method using contextualized capsule networks.The method incorporates the textual contexts into the capsule network to dynamically capture and utilize context information of target entities.Compared with benchmark model,our method achieved competitive results,obtained 86.58% and 93.70% F1 score on the BC5CDR-disease and BC5CDR-chemical datasets respectively.The experimental results show that it improves expression ability of processing rare or unseen entities.(2)Incorporating syntactic information into biomedical named entity recognition using graph convolutional network.Traditional named entity recognition methods understand text as a one-dimensional collection of input vectors,ignoring the correlation and hierarchical information between entities.Syntactic dependency can potentially infer the existence of specific named entities.In this paper,we propose a method incorporating syntactic information in the biomedical named entity recognition using graph convolution neural network.Specifically,each word in the text is regarded as a node,the edges between the nodes are constructed through a syntax tree,and the internal structure information of the entity is captured using a graph convolution neural network.Experiments are conducted onstandard datasets and the results show that the proposed method can effectively capture the entity structure information,with an increase of about 1% F1 value compared with benchmark model.