W3C home > Mailing lists > Public > public-rif-wg@w3.org > December 2005

[Use Case] FZI-1 Ontology Mapping with OWL and Rules

From: Markus Krötzsch <mak@aifb.uni-karlsruhe.de>
Date: Mon, 5 Dec 2005 00:25:46 +0100
To: public-rif-wg@w3.org
Message-Id: <200512050026.00995.mak@aifb.uni-karlsruhe.de>
** Ontology Mapping with OWL and Rules

* Outline

A researcher wants to use the internet to find out about current job 
opportunities at universities and research institutes. Many of the 
institutions she is interested in provide OWL-based ontologies that describe 
available job offers. So, instead of referring to numerous different 
institute websites individually, the user can utilize a specialized search 
engine that provides a customized query interface. There, the user may select 
a general domain of interest (e.g. "universities") and specify URLs of 
additional relevant ontologies. An "ontology" in this setting obviously 
involves both instance data (ABox), i.e. job offers, and schema information 
(TBox) that classify these instances.

In order to formulate the query (e.g. asking for "a post-doc position with 
research topics in knowledge management and natural language processing"), 
the search engine offers a customized user interface to generate queries with 
respect to the local ontology of the search engine (including, e.g., the 
concept of a "post-doc"). Furthermore, users can refer to concepts that are 
specific to any of the target ontologies (which might provide the concept of 
"natural language processing").


Now in order to resolve a query, the search engine first collects the 
requested knowledge bases and information about the relationships between 
them. Such relationships might be given in form of other ontologies: e.g. an 
OWL ontology can specify that the class of "experienced project 
leaders" (occurring in one ontology) only consists of individuals that have a 
relation "leads project" (used in some other ontology) with at least two 
projects. However, additional expressivity will often be required. For an 
example, assume that one institute provides a role "projectLeaderOf" between 
positions and projects, and a role "hasTopic" between projects and research 
topics. In contrast, another institute directly associates offered positions 
with topics, using a role "worksOnTopic." The obvious relationship between 
these two conceptualizations cannot be expressed in OWL-DL, but it can be 
cast into some form of derivation rule. Various rule paradigms could possibly 
be used for this purpose, but, as discussed below, our use case imposes 
certain further restrictions and requirements that go beyond mere 
expressivity, but address the question which specific semantics the rule 
language should have.

Depending on how much expressivity is required to describe some relationship, 
both rule specifications and OWL ontologies could be employed by users to 
describe mappings. This is important, since the mappings are not specifically 
created by the search engine, but are gathered from the web as well. For 
example, institutions might already provide mappings to common domain-specific 
reference ontologies, but mappings can also be generated in a (semi)automated 
fashion, retrieved from some central "mapping repository," or just provided 
by individual users who manually specified certain mappings for their own 
usage. In consequence, mappings are as heterogeneous as ontologies in 
general, and there might be several overlapping mappings between two 
ontologies. Recall that we are dealing with the World Wide Web, which is 
bound to be heterogeneous.


By virtue of a well-designed rule language, the search engine can now proceed 
to answer the query of the user by evaluating it against the combined 
knowledge base of the considered ontologies and the mappings between them. It 
is understood that aggregation of data always imposes large scalability 
challenges so that sophisticated methods of optimization and modularization 
will be needed in the general case. 


* Implications

Alignments are aggregated from many sources on the web, and may be specified 
in OWL (whenever this suffices), or in combination with a more expressive 
rule language. It is thus important that the rule language is semantically 
compatible with OWL, both in a well-understood formal sense and in the 
informal sense that the derived answers to a query reflect the intended 
meaning of the individual specifications.

Furthermore, while added expressivity is a desirable goal for a rule language 
in general, the current use case suggests restriction to a fragment of the 
rule language that is still decidable. This is in keeping with one of the 
design decisions of OWL, namely of specifying a decidable fragment -- OWL DL 
-- as ontology language. We need not reiterate the general benefits of a 
decidable fragment here. Just observe that our use case requires the 
automatic aggregation of heterogeneous semantic specification from the web, 
such that it would be desirable that the respective software can at least 
check whether the aggregated knowledge is consistent or not.


A known rule language that meets these requirements is consituted by so-called 
DL-safe rules, a decidable fragment of SWRL for which an efficient reasoning 
system is available in form of the KAON2 OWL-reasoner [1]. Further details on 
the usage of such rules in alignment are discussed in [2], and an explanation 
of DL-safe rules and their interaction with OWL is found in [3].

[1] http://kaon2.semanticweb.org
[2] Peter Haase and Boris Motik. A mapping system for the integration of 
OWL-DL ontologies. In Proceedings of the ACM-Workshop: Interoperability of 
Heterogeneous Information Systems (IHIS05). November 2005.
[3] Boris Motik, Ulrike Sattler and Rudi Studer. Query Answering for OWL-DL 
with Rules. Journal of Web Semantics 3(1), 2005.


-- 
Markus Krötzsch
Institute AIFB, University of Karlsruhe, D-76128 Karlsruhe
mak@aifb.uni-karlsruhe.de        phone +49 (0)721 608 7362
www.aifb.uni-karlsruhe.de/WBS/     fax +49 (0)721 693  717

Received on Sunday, 4 December 2005 23:26:21 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 2 June 2009 18:33:24 GMT