Skip to main content

Scalable and Complete Ontology Reasoning

Final Report Summary - SCORE (Scalable and Complete Ontology Reasoning)

Ontologies expressed in the W3C standard OWL 2 DL are gradually becoming the basis for many research and industrial strength applications. In such a setting the data are stored in (possibly distributed) databases, while their semantics (schema) are described in a formal way using ontologies. Then, answers to user queries reflect both the stored data as well as the knowledge that has been described in the ontology. Unfortunately, query answering over OWL 2 DL ontologies suffers from very high computational complexity; hence, designing systems that can deliver scalable query answering over OWL 2 DL ontologies on top of billions of data is an extremely difficult task.

The typical approach so far was to restrict the expressivity of the used ontology into one of the known tractable fragments of OWL 2 DL, like OWL 2 RL, for which query answering can be implemented within polynomial time. Unfortunately, there are many applications where the full expressivity of OWL 2 DL is required and for these cases the existing technology is still at a pre-mature level.

The goal of the existing project is to design techniques and provide tools that can deliver scalable query answering over OWL 2 DL ontologies. The project proposed and studied two novel approaches. On the one hand the repairing technique can be used to transform (rewrite/compile) at a pre-processing step the input OWL 2 DL ontology into a less expressive but equivalent one, e.g. transformed into an OWL 2 RL one. Then, a scalable OWL 2 RL system can be used on top of the rewritten ontology to compute the right answers to the user query in an efficient way. The researcher showed that in the general case this is not always possible even if P=NP, however, in many practically relevant cases it is, and for these cases the researcher proposed a resolution-based algorithm. He also designed several rewriting/transformation algorithms for various fragments of OWL 2 DL for which it is always possible to transform an ontology into OWL 2 RL, like the fragments ELHI and Horn-SHIQ. In the project the researcher also showed that OWL 2 RL systems can be used to also answer queries containing existential variables, a feature that goes beyond their capabilities.

On the other hand the researcher developed a second approach to the problem mostly to remedy the cases that no transformation exists. More precisely, instead of fully rewriting the input ontology, the new approach attempts to rewrite only those parts for which this is effectively possible (i.e. a “partial” rewriting) and mark the rest. When a user query regards the former part, then a highly scalable OWL 2 RL reasoner can be used to compute its answers otherwise, if it regards the latter, then one needs to resort to an OWL 2 DL system. Note that the unrepaired part, if non-empty, is expected to be small, hence, calling the OWL 2 DL reasoner should rarely happen in practice. This new hybrid approach overcomes the negative theoretical results that were discovered dueing the first phase because some partial repair always exists.

All of the previous techniques have been implemented in a prototype OWL 2 DL query answering system which is publicly available from the following web site:

As part of the further development of the skills of the researcher and his long-term re-integration, he has also been engaged in the supervision of most of the PhD students of our group, in project proposal writing for securing additional funding, and finally, to some extent, also in undergraduate courses.