Querying RDF Schema Views of Relational Databases

Petrini, J. 2008. Querying RDF Schema Views of Relational Databases. Acta Universitatis Upsaliensis. Uppsala Dissertations from the Faculty of Science and Technology 75. 117 pp. Uppsala. ISBN 978-91-554-7202-3. The amount of data found on the web today and its lack of semantics makes it increasingly harder to retrieve a particular piece of information. With the Resource Description Framework (RDF) every piece of information can be annotated with properties describing its semantics. The higher level language RDF Schema (RDFS) is defined in terms of RDF and provides means to describe classes of RDF resources and properties defined over these classes. Queries over RDFS data can be specified using the standard query language SPARQL. Since the majority of information in the world still resides in relational databases it should be investigated how to view and query their contents as views defined in terms of RDFS meta-data descriptions. However, processing of queries to general RDFS views over relational databases is challenging since the queries and view definitions are complex and the amount of data often is huge. A system, Semantic Web Abridged Relational Databases (SWARD), is developed to enable efficient processing of SPARQL queries to RDFS views of data in existing relational databases. The RDFS views, called universal property views (UPVs), are automatically generated provided a minimum of user input. A UPV is a general RDFS view of a relational database representing both its schema and data. Special attention is devoted to making the UPV represent as much of the relational database semantics as possible, including foreign and composite keys. A general query reduction algorithm, called PARtial evaluation of Queries (PARQ), for queries over complex views, such as UPVs, has been developed. The reduction algorithm is based on the program transformation technique partial evaluation. For UPVs, the PARQ algorithm is shown to elegantly reduce queries dramatically before regular cost-based optimization by a back-end relational DBMS. The results are verified by performance measurements of SPARQL queries to a commercial relational database.