Sebastian Wandelt

Distributed island-based query answering for expressive ontologies

Scalability of reasoning systems is one of the main criteria which will determine the success of Semantic Web systems in the future The focus of recent work is either on (a) systems which rely on in-memory structures or (b) not so expressive ontology languages, which can be dealt with by using database technologies. In this paper we introduce a method to perform query answering for semi-expressive ontologies without the limit of in-memory structures Our main idea is to compute small and characteristic representations of the assertional part of the input ontology Query answering is then more efficiently performed over a reduced set of these small represenations We show that query answering can be distributed in a network of description logic reasoning systems to scale for reasoning Our initial results are encouraging.

Efficient instance retrieval over semi-expressive ontologies

In the last years, the vision of the Semantic Web fostered the interest in reasoning over growing sets of assertional statements in ontologies. Traditional tableau-based reasoning systems have problems to answer queries over large ontological data sets because these reasoning systems are based on efficient use of main memory data structures. Increasing expressivity and worst-case complexity further tighten the memory burden. The purpose of this thesis was to investigate how to release the main memory burden from tableaubased reasoning systems and perform efficient instance checking and instance retrieval over semi-expressive ontologies. The key idea was to reduce instance checking for an individual in an ontology to smaller subsets of relevant axioms. Modularization techniques were introduced and further refined in order to reduce the module size. Instance retrieval performance was addressed by defining similarity criteria over individuals and their modules. Finally, this thesis investigated techniques to preserve modularizations under syntactic ontology updates. For evaluation purposes, experiments on benchmark and real world ontologies were carried out. Modularization techniques gave rise to a distributed implementation for solving instance checking and retrieval problems. The principal conclusion is that the main memory dependency for instance checking and instance retrieval can be released from tableau-based reasoning systems for semi-expressive ontologies in practice.

Scalability of OWL reasoning: role condensates

In the last years, there has been an increasing interest in the performance of reasoning on the Semantic Web in presence of large ABoxes. Traditional reasoners make heavily use of in-memory structures and are therefore not suitable to deal with large ABoxes directly. We propose an approach that, informally speaking, works on a composite representation of the role-part of a SHIQ knowledge base.With respect to conjunctive queries, this helps us to provide a kind of proxy that restricts the set of possible bindings for a variable in advance. Furthermore we can use this proxy to reject several queries with no answer substitution immediately. Most notably our approach is query independent.

Islands and Query Answering for Alchi-ontologies

The vision of the Semantic Web fostered the interest in reasoning over ever larger sets of assertional statements in ontologies. Today, real-world ontologies do not fit into main memory anymore and therefore tableaux-based reasoning systems cannot handle these large ontologies any longer.