Knowledge Extraction For Medical Literature Abstracts

The abstract part of medical literatures contains a wealth of medical knowledge and is an important source of medical knowledge.Extracting medical entity knowledge from medical literature abstract can help medical experts construct medical knowledge graphs and assist doctors to acquire and learn medical knowledge quickly.This paper focuses on the extraction of medical entity knowledge according to the textual characteristics of medical literature abstracts.By combining Bidirectional Long Short-Term Memory Network and Condition Random Field,a medical entity extraction model for medical literature abstracts was designed,and a knowledge extraction tool for medical literature abstracts was realized to help doctors quickly extract medical knowledge such as medical entities.The main work of this thesis:1.Developed a medical knowledge framework for medical literature abstracts,designed a set of corpus labeling criteria for medical literature abstracts,and generated an annotated data set based on the domain knowledge of medical experts.2.Designed a medical entity extraction model for medical literature abstracts,in which the medical entity label prediction methods include three: BiLSTM+Softmax based medical entity label prediction method,BiLSTM+CRF based medical entity label prediction method,and BiLSTM+Attention+CRF based medical entity label prediction method.The medical entity extraction model based on BiLSTM+Attention+CRF was applied to the JNLPBA2004 data set,and the precision,recall and F1 values of the model were analyzed through experiments.The results showed that based on BiLSTM+Attention+CRF medical entity extraction model,the accuracy of knowledge extraction from medical literature abstracted was improved,and the problem of doctors’ inability to acquire TCM knowledge from medical literature abstracted was solved.3.Designed and implemented a knowledge extraction tool for medical literature abstracts.Firstly,medical entity knowledge is extracted from the abstract of medical literature by using the medical entity extraction model,and then the relationship knowledge between medical entities is extracted by combining with the medical knowledge framework.Finally,medical knowledge in the form of RDF is formed.The main contribution of this thesis is the medical entity extraction model for medical literature abstracts.