For link discovery and quality assessments of medical knowledge graphs, we propose a novel neighborhood – based relationship entity embedding algorithm to simultaneously predict both noisy and missing links between medical entities such as diseases and symptoms.