Abstract |
Towards the realization of the vision of the Semantic Web, one of the most significant
tasks to be performed is the transformation of current human-oriented Web
information into machine-processable Web information. In this direction, standards
have been adopted in order to structure the data (XML) and to describe the
semantics of the data (meta-data expressed in RDF).
RDF is a data model and along with the RDF Schema, which defines the vocabulary
of this model, they form a mechanism which provides a formal, machine
processable representation of knowledge. However the nature of world is dynamic
and as the world changes, the knowledge itself, or our view of it, is subject to
changes. Consequently, modeling a dynamic world means encapsulating a mechanism
for updating knowledge.
The algorithms dealing with the incorporation of new knowledge in an ontology
(ontology evolution) often share a rather standard process of dealing with
changes. We acknowledge that this process consists of the specification of the language,
the determination of the allowed update operations, the identification of the
invalidities that could be caused by each such operation, the determination of the
various alternatives to deal with each such invalidity, and, finally, some (manual or
automatic) selection mechanism that allows singling out the “best” of these alternatives.
Unfortunately, most ontology evolution algorithms implement these steps
using a case-based, ad-hoc methodology, which is cumbersome and error-prone.
Knowledge updating is a problem which has been thoroughly examined in the
field of Artificial Intelligence under the term belief change. One key idea in the
belief change field is that an update operation should produce an updated belief
which is as close as possible to the original belief. This approach is often described
as minimal change approach. Trying to define “minimal” a lot of propositions have
been made, among which is the definition of an ordering of the possible update
results.
This work presents a framework for updating knowledge, where both the initial
knowledge and the update are expressed in a special subset of First Order Logic.
Updating is based on a well-formed set of Integrity Constraints on this logic and
a predefined ordering between the possible update results. Using this framework
we apply this updating mechanism to a specific application: the RDF/S language.
We define a model to express RDF language in terms of First Order Logic; an ordering
between possible update results and build optimizations of the framework’s
updating techniques based on RDF’s particular set of integrity constraints.
Through the application of our framework’s techniques on RDF/S we express
how the peculiarities of a specific language (which can be expressed with First Order
Logic) could be used to optimize the proposed framework for the specific case.
On the practical side, we speedup our general-algorithm by developing several,
special per-operation, versions of it, which are also formally equivalent to it. We
also discuss a number of issues raised during the implementation of the algorithm in a real-world environment.
|