Biomedical Documents Classification Based On Hierarchical Attention Mechanism And Graph Neural Network

With the rapid development of Internet information technology,text data has an explosive growth.In the face of the complex text information,how to effectively manage and classify the text data and help users to find the required information quickly,accurately and comprehensively is a very meaningful research topic.Text classification is an important and classical problem in natural language processing and has been widely used in many fields.In the biomedical field,the classification of biomedical texts is a fundamental task,and the classification of biomedical literature helps researchers to obtain target information quickly.The existing text classification methods fail to make full use of the hierarchical semantic structure of biomedical literature,leading to undesirable performance of classification of biomedical documents.According to the above observations,the following research work is completed in this paper:(1)According to the hierarchical structure in biomedical texts,the hierarchical attention mechanism is used to model documents.Specifically,the paper uses three levels of attention mechanism:word to word,sentence to sentence and structure to structure of the abstract.Through combining both local and global contexts in the documents,meaningful representations are obtained.(2)Aiming at the problem of data unbalance which is common in biomedical applications,the cost-sensitive learning approach is used to in our model.In addition,an adaptive method is proposed to learn iteratively the best class cost parameters.(3)In order to futher improve the performance of the classification model,this paper utilizes the graph neural network to capture the grammatical features of the text.Specifically,the dependency parsing tree is used obtain the syntax regularities of sentences,deriving better representations for the biomedical text classification model and improving the performance of biomedical text classification.The results of experiments have proved that the proposed text classification models have achieved good results and are superior to other existing models to some extent.In addition,the adaptive class cost learning method proposed in this paper can be used to solve the class imbalance problem in other tasks/problems.