Updates in the context of ontology-based data management

Abstract

Nowadays, the amount of structured data on the Web has increased dramatically with the adoption of standardised vocabularies and ontologies and the standardisation of SPARQL query language. Ontologies as a conceptual, formal representation of a domain, are used to describe the data and, as a key functionality to provide reasoning capabilities. Both data and ontologies in the realm of the Semantic Web are represented as triples in RDF building a “semantic graph” of facts and axioms. SPARQL Entailment Regimes clearly specify the interplay between SPARQL queries and ontologies, so that as answer to queries implicit triples are returned as well. As structured data have grown over time also thanks to people and agents publishing them adhering to the Linked Open Data principles questions of data management and storage became apparent. As a consequence, we have witnessed a development of be- spoke RDF data management systems called triple stores capable of storing, managing and reasoning over billions of triples. Such triple stores are capable of storing big knowledge bases, for instance, DBpedia, being in the centre of the Linked Open Data cloud. The structured data in the case of DBpedia are extracted from Wikipedia infoboxes via a set of mappings, so-called extractors. DBpedia is a dataset that many applications consume by using SPARQL queries via so-called SPARQL endpoints, which in turn are established on top of triple stores. Nevertheless, the vision of the Semantic Web is to enable “Read/Write” Web of structured data; for this reason, the SPARQL Update language was introduced. Remarkably though, the SPARQL Entailment Regimes specification by the W3C does not define the interplay between the SPARQL updates and ontologies. This dissertation aims to study different update semantics posed over data and ontologies, by taking into account different ontology language and mapping expressiveness respectively. In other words, the aim is to study SPARQL updates in the context of Ontology-Based Data Management (abbr. OBDM). First, we propose different update semantics for triple stores that store ontologies together with data, expressed in the RDFS ontology language. We distinguish therein between update semantics for data versus ontologies, by treating them separately. Then, as a next step, we increase the expressivity of the ontology language by adding class disjointness axioms, where in this setting we propose different update semantics inspired by belief revision that are able to handle inconsistencies in the data. In the end, we investigate different update semantics in the full setting of OBDM, i.e., where also mappings to underlying relational data sources are considered and the ontology language is even more expressive adopting DBpedia as a use case.