CONSTRUCTING A KNOWLEDGE GRAPH EMPLOYING MULTIPLE SUBGRAPHS AND A LINKING LAYER INCLUDING MULTIPLE LINKING NODES

A graph-based data storage and retrieval system in which multiple subgraphs representing respective datasets in different namespaces are interconnected via a linking or "canonical" layer. Datasets represented by subgraphs in different namespaces may pertain to a particular information domain (e.g., the health care domain), and may include heterogeneous datasets. The canonical layer provides for a substantial reduction of graph complexity required to interconnect corresponding nodes in different subgraphs, which in turn offers advantages as the number of subgraphs (and the number of corresponding nodes in different subgraphs) increases for the particular domain(s) of interest. Examples of such advantages include reductions in data storage and retrieval times, and enhanced query/search efficacy, discovery of relationships in different parts of the system, ability to infer relationships in different parts of the system, and ability to train data models for natural language processing (NLP) and other purposes based on information extracted from the system.