Construction And Application Of Knowledge Graph In Epilepsy

Epilepsy is a common brain disease.The symptoms will bring great trouble to the patient’s life.However,it is difficult for the public to understand the scientific knowledge about epilepsy due to limited opportunities and complex introduction.The most important feature of the knowledge graph is that it abstracts the complex relations in the real world into a directed graph composed of nodes and edges.With the help of the epilepsy knowledge graph,the connection of things becomes intuitive.People can more easily obtain epilepsy-related information.For example,query the connection and commonality between different types of epilepsy.The key steps of knowledge graph construction are named entity recognition and relation extraction.However,neither task has annotated datasets in the epilepsy domain.This leads to the inability of mainstream approaches to achieve good results.In addition,mainstream approaches generally model the relation extraction task as a classification task,which can only extract specific relations,resulting in incomplete extraction results.Therefore,we propose a named entity recognition and relation extraction model that does not rely on epilepsy domain annotated datasets and use this model to develop a knowledge graph construction system in epilepsy.This system provides users with a knowledge graph construction service from papers.Specifically,our works are as follow:1)To solve the problem that the named entity recognition model lacks a training set in the epilepsy domain,we propose a model for the named entity recognition task in the epilepsy domain based on domain adversarial neural network and multi-task learning.Both strategies improve the generalization ability of the model by affecting the LSTM layer and the embedding layer.The proposed model achieves an F1 value of 0.8348 on the self-labeled test set,proving the effectiveness of the improvement.2)To solve the problem of lack of training sets and limited types of relations in the relation extraction task,we introduce prior rules and a discriminator after the open relation extraction model.The quality of relation triples is improved by filtering relation triples from several aspects.The experimental results show that the quality of extracted triples has significant improvement,and the accuracy of the improved model in the sampling evaluation is up to 42%.3)We design and develop the epilepsy knowledge graph construction system in B/S mode based on the above models.The system can convert uploaded papers and extract named entities and relation triples from papers.It can provide users with the service of constructing a knowledge graph with one click.And the system also provides the demo page and the search function for knowledge graphs so that users can easily browse the knowledge graph generated by the system.