Information Technology Reference
In-Depth Information
automatic solution less feasible. In contrast, semantic approaches [4,25,23,26,1,18]
are based on an explicit modeling of ontologies which capture a specific domain
knowledge in a generic way using concepts that are commonly agreed and under-
stood [9]. As a consequence, machines become more able to interpret data and
documents that are annotated with concepts from ontologies requiring less or no
human intervention. In general, research communities distinguish between two
types of metadata, i.e. ontological concepts and annotations encoding references
to concepts.
In terms of ontological concepts, we described what can be seen as a rather
closed ontology derived from a given methodology. Such an ontology can be
considered nearly complete with respect to the universe of discussion. However,
we did not focus on the semi-automatic creation of ontologies [15,6,12] such as
generating ontologies from natural language text or automatic reuse of existing
ontologies, e.g. based on the context of a document [22]. In fact, reusing public
ontologies, e.g. TAP [7] or others, are less applicable in context of our work
as design methodologies in an enterprise context rather describe proprietary
concepts (e.g. data model) that cannot be necessarily expressed with public or
community-contributed [24] ontologies.
In terms of annotations, existing approaches can be first categorized into
degrees of automation, i.e. fully-automated, semi-automated [9,15] or manual
[14,19]. Second, the annotation level of detail depends on the type of represen-
tation ranging from e.g. entire service descriptions, operation or input/ouptut
parameters, non-functional requirements etc. Third, annotations can be used for
different fields of applications, e.g. human or automatic service discovery [17],
invocation, composition [22] etc. In this work, we proposed a fully-automated
approach to annotate a large set of Enterprise Services based on their signa-
tures, i.e. interface and operation names. Ultimately, we currently see the main
purpose of the service annotation approach in improving the service discovery
specifically for business users and non-professional developers. In this context,
using automata can more precisely resolve disambiguities by determining correct
concepts based on their expected position within the respective signature. As a
result, we can increase the accuracy of generated annotations compared to other
approaches, e.g. using similarity functions [7] (SemTag), natural phrase process-
ing [9] (via SMES) or detecting association of concepts by confidence level [15].
Note, this is particularly feasible as we only focus on a small portion of explicitly
defined text, i.e. Enterprise Service signatures, rather than a large body of text,
i.e. Web pages.
For reasons of simplicity, we used RDF(/S) over OWL-S or WSMO to store
our ontology and annotations. We consider RDF(/S) sucient for our purpose
to represent additional knowledge as we do not require sophisticated logical
reasoning. The results of the annotation could, e.g., be stored in SAWSDL [25,23]
format, where annotations are directly added to WSDL. However, at this point
we are not entitled to change Enterprise Service descriptions, and decided to
keep annotations independent from any specific Web Service standard.
