Constructing a Multi-scale Medical Knowledge Graph from Electronic Medical Records

Knowledge graphs play a crucial role in the medical field. Most existing knowledge graphs are manually created by experts or extracted from medical encyclopedias, resulting in the omission of valuable knowledge from medical clinical practice. Entities like diseases and symptoms in medicine exist at different levels, but current knowledge graphs fail to handle the induction and integration of this multi-scale information effectively. In our study, we constructed a knowledge graph that better aligns with real clinical data and effectively integrates multi-scale medical information by performing data preparation, medical entity extraction, negation handling, relation extraction, and graph cleaning. The reliability and rationality of the knowledge graph have been verified through subjective and objective assessments.