A framework for integrating and transforming between ontologies and relational databases

• Main Author: Al Khuzayem, Lama
• Other Authors: McBrien, Peter ; Russo, Alessandra
• Published: Imperial College London 2017
• Subjects: 004
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.754708

Bridging the gap between ontologies, expressed in the Web Ontology Language (OWL), and relational databases is a necessity for realising the Semantic Web vision. Relational databases are considered a good solution for storing and processing ontologies with a large amount of data. Moreover, the vast majority of current websites store data in relational databases, and therefore being able to generate ontologies from such databases is important to support the development of the Semantic Web. Most of the work concerning this topic has either (1) extracted an OWL ontology from an existing relational database that represents as exactly as possible the relational schema, using a limited range of OWL modelling constructs, or (2) extracted a relational database from an existing OWL ontology, that represents as much as possible the OWL ontology. By way of contrast, this thesis proposes a general framework for transforming and mapping between ontologies and databases, via an intermediate low-level Hyper-graph Data Model. The transformation between relational and OWL schemas is expressed using directional Both-As-View mappings, allowing a precise definition of the equivalence between the two schemas, hence data can be mapped back and forth between them. In particular, for a given OWL ontology, we interpret the expressive axioms either as triggers, conforming to the Open-World Assumption, that performs a forward-chaining materialisation of inferred data, or as constraints, conforming to the Closed-World Assumption, that performs a consistency checking. With regards to extracting ontologies from relational databases, we transform a relational database into an exact OWL ontology, then enhance it with rich OWL 2 axioms, using a combination of schema and data analysis. We then apply machine learning algorithms to rank the suggested axioms based on past users’ relevance. A proof-of-concept tool, OWLRel, has been implemented, and a number of well-known ontologies and databases have been used to evaluate the approach and the OWLRel tool.