Automated Knowledge Graph Construction for Healthcare Domain

This research seeks to optimize the process of identifying correlations in common and high severity diseases via the fusion of knowledge graphs and deep learning artificial intelligence. Knowledge graphs can be complicated to construct and resource-intensive, alternatively, knowledge graphs can be seen to legitimize correlation incidence and better explain AI outputs. We propose automation of knowledge graph construction from identifying significant text frequency relations within established knowledge base document structures to identifying inter-feature relations and creating a novel approach for artificial intelligence and machine-learning feature extraction and feature selection. Our knowledge graph construction exploits the structured World Health Organization (WHO) International Classification of Disease (ICD) code chapters, which are specific to a single organ system of the human body. A sorted vector of text-to-chapter frequencies enables Wilcox Rank significance tests to determine the most related features.