Research On Entity Linking Method For Clinical Electronic Medical Records

Real-world data is one of the important data sources for medical research,and clinical electronic medical record data is an important part of real-world medical data.Due to the wide variety of entities in clinical electronic medical records and the variety of entity expressions,the first step when using clinical electronic medical records for research is to standardize and normalize the entities in clinical electronic medical records.However,the number of clinical electronic medical records is huge.Manual specification is time-consuming and laborious,so an automated electronic medical record data normalization method is needed.Entity linking can solve the problem of entity standardization,but there is relatively little research on entity linking in the medical field.This article studies the concept linking method of the disease entity mentioned in the clinical electronic medical record.Combining the two steps of entity linking,two types of methods are proposed respectively.The main research work includes the following two aspects:(1)Firstly,in order to carry out the research of entity linking method,a content expansion method of disease knowledge base based on web crawler is proposed.By acquiring and structuring medical Web resources such as Encyclopedia of Medicine,this article is used to supplement the definition,etiology,examination and other relevant information in the disease knowledge base,forming a disease term classification knowledge base with more than 50,000 entities.Secondly,on this basis,for the candidate entity generation step,a disease candidate entity generation method based on text classification is proposed.The algorithm generates candidate entities based on the clinical characteristics of medical data and using the characteristics of the disease department.For the generation of candidate entities for a single disease entity,the single entity candidate entity generation method is adopted,and the disease entities are vectorized word by word,and then the deep learning classification model is used to realize the department classification of the single disease entity.For the problem of generating candidate entities for disease entities with associated texts,the method of generating associated candidates is used to vectorize the entity name and the associated text of the entity,and a deep learning classification model that combines two vector features is constructed to achieve association.Department classification of sexual disease entity.The experiments on the constructed data set show that the above algorithm has certain applicability,and the experimental structures of the above two methods are analyzed and compared.(2)According to the candidate entity ranking linking step,a disease candidate entity linking ranking method based on similarity calculation is proposed.This method uses six similarity calculation algorithms: edit distance,Jaro-Winkler similarity,cosine similarity,Euclidean distance,Jaccard coefficient,and Dice coefficient to calculate the similarity characteristics between two characters,and uses logistic regression to classify The model outputs the probability that each candidate entity belongs to the linkable class,and finally sorts out according to the probability to filter out the knowledge base entries that can be linked.In order to verify the effectiveness of the above entity linking model,the disease alias data in the knowledge base is selected for testing,and the final structure is evaluated and analyzed.