During the prototype construction we have been discover several
problems.
Mapping OpenCyc to OWL
Opencyc owl version lacks the expressiveness needed to satisfy
a whole and covered semantic model. It does not uses description
logic, and it does not define “domain” and “range”
on the ontology properties too.
Analyzing the problem in depth, Opencyc version has been built
without a through mapping process between CycL and OWL languages.
An OWL Class corresponds to a CycL concept, whereas an OWL “ObjectTypeProperty”
corresponds to a CycL predicate. The main relationships defined
in the RDFS specification have been mapped. For instance, #$isa
and #$genls has been converted into “subClassOf” properties.
Also, many “subPropertyOf” relationships have been mapped
(figure [12]).
OWL Class: #$Person
(#$arg1Isa ?X #$Person)
Properties: placeOfWork , boyfriend, nativeLanguage, etc..
#$Person --> http://www.cyc.com/2003/04/01/cyc/#Person
boyfriend -->http://www.cyc.com/2003/04/01/cyc/#boyfriend
Figure 12 - #$boyfriend(PERSON MALEHUMAN): CycL predicate
converted to OWL property into a Jena Persistent Model.
But, what happen when a predicate has an arity greater than one?
A predicate only can correspond to OWL ObjectProperty, but CycL
allows a limited number of arguments and provides the mechanisms
to define more than one predicate arguments. However, in the OWL
file, if a predicate has several arguments, there not exists a formal
specification to allocate arguments. Also, the range and domain
properties have not been defined. As it can be seen in figure [13],
the “BorderOf” predicate does not have these properties.
Figure 13- #$borderOf(BODER REGION) CycL predicate
converted to OWL property into a Jena Persistent Model.
Low performance
The tests run on semantic web frameworks
have highlighted problems when handling large amount of data. Many
of these problems are due to inference execution over the underlying
persistent model. In detail, when a query is dispatched the reasoner
planning the execution to retrieval all inferred data from the database.
Interface problems
The slor interface needs a hard handle
to integrate the semantic concepts and rules within metadata lom
records.
In order to solve these problems we had considered new web development
trends. AJAX (witch stands for Asynchronous Java Script and XML)
is a newfangled technique that provides a new scenario for the interaction
between web client and server. AJAX increases the web page’s
interactivity, speed and usability. This technology makes web pages
more dynamic, disabling the post back actions of the current web
technologies (such as ASP.NET or Java Server Faces) and providing,
at the same time, a new asynchronous communication based on XML
messages between server and client. Hence a web client don’t
need to send a dynamic web page and wait to receive it again (by
means of XMLHttpRequest object).
AJAX provides a set of encapsulated XML messages over HTTP protocol,
Figure [16] shows a comparison between web application models. A
detailed discussion about AJAX interaction is out of the scope of
these technical notes, however interested readers can consult (Garrett,
2005).
Figure 14 - SLOR web interface
We have developed a simple AJAX component to test the benefits of
this technology in SLOR. Similar to Intellisense technology, this
component helps to write the semantic expressions within learning
object metadata records. It is not necessary to know about an ontology
and all its individuals, class or properties. As it can be seen
in the figure [15], the AJAX component shows a list that contains
all individuals of the country class related with the property “situatedIn”
making used of the TGN ontology.
Figure 15 - SLOR AJAX Component
Figure 16 - Comparison between web interaction models.
(Source Garret, 2005)
Hard Work
We can not leave the research area of automated
metadata creation, if we want to create a usable repository. Each
learning object specification provides a big set of fields to describe
a learning object in different context. To fill and complete a whole
description is a hard job, although there exits several tools to
create metadata records. Studies as (Philippe Vidal and Michael
Meire, 2005) shows how can uses an automated assistant that auto
complete some obvious and usual fields.
|