Formalizing Relational Databases as OWL Ontologies

While the World Wide Web provides a platform for sharing information, using it to answer sophisticated questions involving significant background knowledge is not currently possible. The principal reason behind this limitation is that digital information, while machine processable, is not machine understandable. In contrast, the emerging Semantic Web leverages current web architecture and additionally provides the means by which explicit semantics can be assigned to raw data through ontologies that explicitly describe and relate objects using formal, machine understandable representations. Widespread adoption of the Semantic Web depends significantly on an increase in the number of ontologies spanning different domains of knowledge. Manually designing ontologies, however, is tedious, acting as a knowledge acquisition bottleneck.;DBOwlizer goes beyond other existing approaches by: i) representing mappings using standardized Semantic Web languages, thereby avoiding the use of application specific non-standard mapping languages, ii) facilitating the modification and reuse of mapping heuristics, iii) enabling the comparison of heterogeneous heuristics that map relational databases to ontologies, iv) automatically generating decidable database-to-ontology mappings for more expressive ontologies, v) separating domain knowledge from model knowledge in the output ontology, and vi) mapping semantics from database views, including a pattern to characterize aggregation functions.;We anticipate that this work will help to expose the rich content currently stored in thousands of databases in a way that users, in particular scientists, can be more effective in retrieving information that supports their scientific endeavours.;Keywords Semantic web, knowledge representation, ontology design, ontology engineering, expressive ontologies, data integration, reverse engineering, question answering.;In this thesis, we present DBOwlizer, a novel framework for the automatic design of expressive ontologies from (normalized) relational databases. This approach extracts and accurately represents the information found in a database by examining its composition. DBOwlizer allows one to: i) represent mappings between a relational database and an OWL ontology, ii) automatically generate such mappings using rule-based heuristics and iii) execute these mappings to generate a populated OWL ontology.