Research On Medical Entity Extraction Based On Supervised Learning

The entity extraction is to recognize entities that have special meaning in the text,and the basis of structuring the text.Entity extration is used in many fields as an automatic tools to access important information quickly.It is already quite mature in the area of natural language processing.With the dramatic increase of electronic medical texts,the applications of medical entity extraction have been paid more and more attention to in the medical field.However,for the terminology in the medical field,the accuracy of generic entity extraction is not high.This thesis has studied methods of how to improve the accuracy of the medical entity extraction based on terms.The main methods are the combination of terms and rules and the supervised learning methods.It focuses on the choice of machine learning models and the selection of feature set,so as to achieve the more efficient result of medical entity extraction.In this thesis,methods of terms,rule matching,SVM(Support Vector Machines)and CRF(Conditional Random Fields)are applied on medical entity extraction.This thesis uses two comparative experiments to find the most suitable method of medical entity extraction.On one hand,the methods of terms and rule matching are compared.On the other hand,the supervised learning methods of SVM and CRF are compared by selecting feature and constructing feature template.The experiment results are evaluated by precision,recall,F-score and accuracy and show that the method based on CRF is the most effective on medical entity extraction.