An Efficient Schema Extraction Technique for Graph Databases

Abstract

Predominantly Relational Database Management System (RDBMS) is the major storage system deployed in health-care information systems. Even though it is widely used and remarkably mature, data with high degree of relationships levies a high performance toll on RDBMS. Heavy annotation of health-care data with relationships makes its storage suitable for specialized data models like graph databases or other considerably young NoSQL datastores. Each storage system has its own pros ans cons, the new NoSQL datastores cannot be considered the successors of the hugely established relational model. But storage mechanisms’ true performance can be harnessed by using these models side by side, employing each model’s specialty to store partial data in individual model. This approach needs connectivity and coordination among adopted datastores for precise data integration. Knowledge of schema of each participating datastore is essential for data integration, which the NoSQL datastores do not possess. In this proposal an approach to extract schema from a running graph datastore (namely Neo4j) and its storage in a universal format (Datalog) has been proposed. The methodology employs universal approach for schema extraction by exploiting graph data exported into XML (GraphML) which is supported by most of the NoSQL datastores, hence making the approach generic. Further, algorithms are presented on handling and formulating user queries on the graph data utilizing the extracted schema. The procedure has been illustrated with the help of a medical database - EVD (Ebola Virus Disease). The proposed work is efficiently extracting schema from Neo4j but in order to include other database classes schema extraction technique will have to be slightly modified.