Ontology Learning


Within the last decade, ontologies have emerged as a powerful standard for representing computer-tractable knowledge. Amongst others, the Semantic Web initiative is one of the driving forces behind this development. Current ontology languages like OWL-DL have stable, well-founded and expressive semantics using Description Logics as underlying knowledge representation standard. While the benefits of using ontologies are now well understood, their creation and maintenance is still a burdensome task. Knowledge engineers need more powerful tool support to evolve and manage ontologies.


One of the keys for designing learning algorithms in Description Logics are refinement operators. They allow for an efficient traversal of the subsumption hierarchy of concepts. One way to assess the suitability of a refinement operator for learning algorithms is to look at its properties. We analyse the properties (completeness, weak completeness, properness, redundancy, finiteness, minimality, and their combinations) and show theoretical limitations.
Learning algorithms can be designed by combining a refinement operator with a a search heuristic. We are interested in designing operators which are close to the best one can hope for and combine them with intelligent search heuristics.
As a second approach we investigate the use of Genetic Programming (GP) to solve the learning problem in Description Logics. In particular, we are interested in hybrid technologies combining the evolutionary inspired GP with refinement operator based search.