Research On Medical Entity Semantic Relation Extraction Method For Medicine Instructions

Entity relationship extraction is a branch of information extraction field,this topic researchs how to extract structured knowledge from unstructured text information,and uses semantic information such as natural language processing,machine learning and data mining to facilitate retrieval and further application.Field-independent entity relation research is booming and has achieved good results.But the exploration of the field of medical expertise is just beginning.The medicine instruction is a kind of guide to help selecting medicine.The text of a medicine instruction includes a large number of medical entities,extracting the medical entity relationship from medicine instructions provides a data base for the medical industry as well as great application value.The research on medical entity sematic relation extraction is regarded as the classification problem under the background of a relatively complete relationship system.Considering the specificity of the coreference relation,it is solved by coreference resolution based method.For the medicine instruction corpus,supervised learning methods have been used,the SVM and Naive Bayes classification models are used to predict the relation categories,using the methods of supervised learning with the entity type,the relative position,the upper and lower information and the clue information.Supervised learning methods perform well and made the ideal results.Supervised learning methods require scales of training samples,this paper proposes semi-supervised learning methods for this shortcoming,using a small amount of training corpus of different sizes as the input for self-training algorithm to predict relation categories.The experimental results verify the feasibility of the semi-supervised method for medical entity sematic relation extraction,and show that the semi-supervised learning method has more advantages in the sample utilization than the supervised learning method.