Research On Information Extraction Methods Of Antibiotic Resistance From Biomedical Texts

The situation of antibiotic resistance is severe,and the development of antibiotics is faced with unprecedented challenges.Understanding the mechanism of antibiotic resistance is crucial for tracking the spread of drug resistance,optimizing treatment methods and developing new drugs.A large number of research results about antibiotic resistance exist in biomedical literature.Adopting text mining methods to automatically obtain information from the text effectively reduces the cost of knowledge acquisition and improves work efficiency.In this thesis,antibiotic resistance information extraction is modeled as a biomedical event extraction task,in order to automatically obtain the information of antibiotic resistance mechanism from biomedical text.Biomedical event extraction mainly consists of two sub tasks:biomedical event trigger identification and argument detection.Specifically,biomedical event trigger identification is the premise of biomedical event extraction task,which is the focus of this thesis.At the same time,there are many biomedical knowledge bases that can be used for knowledge retrieval.How to effectively reuse knowledge is also worth exploring.For the research of antibiotic resistance information extraction,this thesis first carries out biomedical event trigger identification task,and then conducts argument detection task.The main work and contributions of this thesis are as follows:(1)Annotation for dataset.This thesis models the information extraction task of antibiotic resistance from biomedical text as the biomedical event extraction task,and labels the corresponding dataset for antibiotic bacterial event extraction(Antibiotic Bacteria Event Extraction,ABEE).In this thesis,according to the classification criteria set by CARD database for the mechanism of antibiotic resistance,we downloaded and manually annotated 1960 medical literature abstracts on bacteria and antibiotic resistance from PubMed database.ABEE includes seven types of antibiotic resistance mechanisms and four biomedical entities,including antibiotics,bacteria,genes and proteins.(2)This thesis proposed a few-shot learning based framework for antibiotic resistance trigger identification.Most biomedical event extraction methods are based on deep learning,and the performance of the model is highly dependent on the size of the training data set,while large-scale training data sets are usually not easy to obtain.Therefore,from the perspective of the scarcity of training corpus in specific fields,this thesis models the trigger identification of biomedical event as a few-shot learning problem,proposes a framework of antibiotic resistance triggers identification based on few-shot learning,and uses self-attention mechanism to effectively fuse external knowledge.(3)This thesis proposed a Bi-LSTM+CRF based framework for argument detection of antibiotic resistance.Biomedical event extraction mainly includes two subtasks:biomedical event trigger identification and argument detection.In order to complete the whole information extraction process of antibiotic resistance mechanism,this thesis adopts a deep learning framework based on bidirectional long-term and short-term memory network and conditional random field for antibiotic resistance event argument detection.This thesis proposed methods to model the information extraction for resistance mechanism of bacteria and antibiotic as a biomedical event extraction task.The results of experiments based on real data demonstrate the effectiveness of the methods.Biomedical event extraction can automatically identify the antibiotic resistance events distributed in biomedical texts,which provides a certain research basis for the follow-up study of antibiotic resistance.