Deep Neural Network Based on Translation Model for Diabetes Knowledge Graph

Knowledge base completion (KBC) can predict new facts according to existed facts in knowledge base. However, most work in KBC is limited to simple neural networks and usually focuses on commonsense knowledge base where data resources are from public websites. Medical records related to diabetes from hospital showed potential advantages in both quality and quantity, which is significant to be mined for preventing or delaying diabetes and its complications. Thus, this paper designs and constructs Diabetes Knowledge Graph from electronic medical records in Shanghai Ruijing Hospital, and proposes a deep neural network based on translation model for completion in the diabetes knowledge graph. We manually evaluate our trained model's ability in accuracy, recall and F1-scores, finding that it has a good stability and performance in diabetes knowledge base completion.