Research On Medical Entity And Relation Extraction Based On Joint Learning

With the development of artificial intelligence,informationization and automation of medical treatment have attracted wide attention.In the healthcare and medical domains,compared with the texts in the public medical field,the medicine instructions are usually well-written,authoritative,and comprehensive.Though they have relatively small data scale,but contain a large number of professional terms and semantic information,and hold obvious characteristics in the medical field.At present,most researches in the healthcare and medical domains are focused on English datasets.Due to the particularity of Chinese syntax and grammar,the research in the Chinese medical field is more complex and difficult.In this paper,the dataset of medicine Chinese instruction manual are independently constructed based on the antibacterial medicine instruction manual.At present,the research in the healthcare and medical domains mainly focuses on named entity recognition and relationship extraction.In the past work,most of them have adopted the pipeline model: first,named entity recognition,then relation extraction.This kind of model ignores the connection between tasks and causes error propagation.At the same time,it needs a lot of manual operations.In view of the drawbacks of pipeline model,this paper introduces joint learning: the convolutional neural network is combined with support vector machine and conditional random field to build a joint neural network model.On the basis of this model,the entity classification and relationship extraction are studied jointly in the way of parameter sharing.In the further experiments,a new joint model is built based on bi-directional long short-term memory neural network.The model can completely extract all entities and relationships without relying on any artificial features or NLP tools.And meanwhile,self-attention mechanism is added to the joint model to learn the dependency between entities.The joint model based on the convolutional neural network tries the methods of task joint,model joint and feature joint to experiment on the dataset of medicine Chinese instruction,and achieves good results.The joint model based on the bi-directional long short-term memory neural network has achieved the optimalresults in both the Chinese dataset and the open English dataset.The experiment proves that joint learning is very effective.