Research On Biomedical Event Extraction Based On Domain Semantic Information

As of 2019,more than 30 million biomedical literatures have been included in the Pub Med database,and the effective biomedical information excavated from these texts for disease diagnosis,new drug research and development and etc has become a research hotspot,and biomedical event extraction is its focus.The main tasks of biomedical event extraction include event trigger detection and event argument recognition.The current related research has the problems of insufficient mining of text semantic features and insufficient annotation of data samples.In view of the problem that the semantic features of biomedical text are not fully excavated in trigger detection,this paper proposed a Learning Key Semantic Information from Multiple Perspectives Method(LKSIMP),which first uses multi-core convolution and full connection layer to learn semantic features from multiple perspectives of text,so as to obtain more semantic information.Secondly,to obtain the important semantic features of trigger detection,this paper proposed a Learn Key Semantic Information Method(LKSI),which determined the importance of word semantics in trigger detection by calculating key semantic information scores.Finally,the LKSIMP-BiLSTM and LKSIMP-TreeLSTM models are constructed according to temporal features of sentences and the dependency tree structure of sentences for LKSIMP.The experimental results on the GE11 and GE09 data sets show that the LKSFMP-BiLSTM model is optimal compared with the accuracy and F1 score of other trigger detection models.In view of the problem that the semantic features in the biomedical field are too abstract and the data samples are insufficiently labeled,this paper introduced the external semantic knowledge to the biomedical text for event argument detection,and proposed a Learning Semantic Information based on Domain Knowledge Method(LSIDK).This method first proposed the Average Concept method and the Similarity based on Concept method to construct the biomedical concept vector separately,to extract the semantic information of gene functions related to the entities in the text from external semantic knowledge.To integrate external knowledge with knowledge in biomedical texts,we proposed a Conceptual Context Tree-LSTM Model(CC Tree-LSTM).The experimental results on the GE11 and GE09 data sets show that the CC Tree-LSTM model of biomedical vector constructed using the Average Concept method is optimal compared to the F1 score of other event argument detection models.