Mapping Relational Databases to Linked Data

To live up to its promise of web-scale data integration, the Semantic Web will have to include the content of existing relational databases. One study determined that there is 500 times as much data in the hidden or deep web as there is in crawlable, indexable web pages; most of that hidden data is stored in relational databases [79]. Starting with a 2007 workshop, titled “RDF Access to Relational Databases”1, the W3C sponsored a series of activities to address this issue. At that workshop, the acronym, RDB2RDF, Relational Database to Resource Description Framework, was coined. In September 2012, these activities culminated in the ratification of two W3C standards, colloquially known as Direct Mapping [43] and R2RML [165]. By design, both these standards avoid any content that speaks about implementation, directly or indirectly. The standards concern is syntactic transformation of the contents of rows in relational tables to RDF. The R2RML language includes statements that specify which columns and tables are mapped to properties and classes of a domain ontology. Thus, the language empowers a developer to examine the contents of a relational database and write a mapping specification. For relational databases with large database schema, the manual development of a mapping is a commensurately large undertaking. Thus, a standard direct mapping is defined; that is an automatic mapping of the relational data to an RDF graph reflecting the structure of the database schema. URIs are automatically generated from the names of database schema elements.