Research On Knowledge Graph Construction And Mining For Pulmonary Diseases

Pulmonary diseases affect the working ability and quality of life of patients.With the progress of the disease,it seriously endangers human health and brings huge losses to social production and economy,which is also an urgent problem to be solved in the medical field.With the development of medical informatization,medical data are increasing day by day.Knowledge graph has become a hot issue of common concern because it can connect scattered and trivial information to realize knowledge integration.It is of great significance to construct the medical knowledge graph.It can extract medical knowledge from the mass data to manage,share and apply it reasonably and efficiently.At present,there are many researches on the construction methods of knowledge graph,but there are many problems such as low efficiency,many restrictions and poor expansibility.At the same time,there is a lack of research on the construction and mining of knowledge graph for pulmonary diseases.In view of the above problems,the paper explores the construction and mining methods of knowledge graph of pulmonary diseases,so as to support the auxiliary diagnosis and prediction of pulmonary diseases.The main work of this paper is as follows:(1)We construct the knowledge graph of pulmonary diseases,which accurately described,clearly layered and effectively visualized the relationship between disease symptoms.The heterogeneous problem caused by different data sources is avoided.We use label consistency to evaluate the constructed knowledge graph of pulmonary disease,which proves the validity of the knowledge graph and provides support for subsequent mining and prediction tasks.(2)We propose a link prediction method for the knowledge graph of pulmonary diseases(LPP).Firstly,the knowledge of pulmonary diseases is quantified and sends to the gated recurrent unit neural network for full learning in the form of triples.At the same time,we introduce the attentional mechanism to express the correlation and finally realize the prediction of pulmonary diseases.The experimental results show that the LPP method which combines the gated recurrent unit neural network with the attentional mechanism is effective,which not only improves the accuracy of the prediction,but also improves the stability of the model.(3)We propose a balanced probability distribution algorithm based on feature and instance transfer(BPD).In this method,cascaded transfer learning based on the instance is used to obtain the instance close to the target domain,and then the co-occurrence feature of the source domain and the target domain is obtained by using the cross-domain feature filtering algorithm.The learning transfer from multi-source domain to target domain is realized by instances and co-occurrence features,the classification model of target domain is constructed,and the generalization performance of the model is further improved by elastic network.Experimental results show that BPD method has the best predictive effect of pulmonary disease.