Construction of heterogeneous medical knowledge graph from electronic health records

Abstract Knowledge graph (KG) is a knowledge organization that enables the users to quickly and accurately query the information required. It is stored in the form of triples. It finds its application in various fields of enterprises, academics, medical etc. In this paper, Medical Knowledge Graph is constructed from Electronic Health Records (EHR) that maps the relationships between the entities of patients, disease and drugs. The heterogeneous graph is constructed using different entities derived from distinct datasets and the information is extracted in the form of queries among the wide between the quantity demand and the wide variety of variables. The steps for construction of Medical Knowledge Graph are data collection, Named entity recognition, entity normalization, entity ranking, and Graph Neural Networks. The hybrid approach of Bi-directional Long Short-Term Memory Multi-Head Attention Conditional Random Field (BILSTM-MULATT-CRF) is used for Named Entity Recognition and the result of Precision 91.4%, Recall 90.15% and F1 score 90.77% is obtained.