Use case collection

The following use-cases for a Web-ontology language were presented during the SIG meeting. (These are the rough notes taken during the presentations by Frank van Harmelen, so all mistakes in these notes are his.

The following use-cases were presented:

Computer System Management (IBM Rome)
Multi-media generation (CWI Amsterdam)
Knowledge management and portal applications (AIFB Karlsruhe)
Document annotation and retrieval (Prague)

A final discussion concerned the introduction of reification into an ontology language.

Computer system management

Guido Vetere from IBM Labs Rome presented work they have been doing since 1996 on a computer system management ontology, which describes computer systems, problems and failures. The ontology was modelled in a Description Logic style. This turned out to be easy. There was no strong commitment to Description Logics, and their full power was not really fully exploited; for browsing this was not necessary, but the Description Logics were useful for reasoning. The ontology modelling took as in put the "Common Information Model" (see http://www.dmtf.org/standards/standard_cim.php), developed by large IT producers (Microsoft, Intel, IBM), with 300 concepts on IT products and processes. This fitted well into a Description Logic. A problem for users was how to find relevant concepts without browsing. This is now done via attaching to WordNet. A new problem raised in this approach was that current ontology languages lack mechanisms for associating lexical descriptions with concepts. A notion of "linguistic signs" is needed because different user communities have different such "signs". The group is currently not using any of the publicly available ontology languages, but has programmed the application in a home-grown datamodel with a private API.

Multi-media generation

Jacco van Ossenbruggen of CWI Amsterdam presented the groups work on multi-media presentation generation: their intended scenario is as follows: in response to a query to a database, a user gets back a set of media-items; the challenge is then to combine these items in a coherent multi-media presentation that answers the query. For this, one need lots of information about these multi-media-items and their relations. Some of the expressivity requirements are:

the need for different ontologies at different levels, eg: media-specific, domain/content-specific, presentation-specific;
the need to reason about all these ontologies combined. The reasoning they need combines reasoning at class and instance level. No subsumption style reasoning seems required.
the need to use large existing ontologies, but typically the only need parts of these large ontologies. Thus, there is a requirement for some information-hiding/modularity concept;
these different existing ontologies are also likely to come from different communities, so there is a need syntactic translations.
Finally, the group also wants to re-use the “input metadata” from the databases and include them into the output presentations. Again, this requires mapping and import mechanisms.

Currently all the work of the group is encoded into home-grown ad hoc technical solutions. However, they are keen to move to a more clean declarative representation.

Knowledge-management and portal applications

The AIFB group of the Univesity of Karlsruhe presented some of their work in ontologies for knowledge-management applications and portal building:

They constructed a skills management ontology for Swiss Life as part of the On-To-Knowledge project. This ontology is intended to be used for gap analysis (= missing skills in the organisation), and expert-finding.
They constructed a ontology of tourism information (1200 concepts, 300 relations). This was a large multi-person effort, and is used for portal construction, in particular to support query-answering
The ontology for the Ontoweb portal was built to support a similar task
Another ontology supports their own AIFB portal with access for different navigation structures

Some of their conclusions from this work are:

There is a need for both schema-queries and instance queries, as well as combinations of these two;
keep it simple (e.g. required for Swiss Life), possibly with more complicated layers,
Until now, they have only had a need for named concepts only. They have found no use for the "anonymous classes" or "class constructor" expressions as found in Description Logics
They have a need to be able to generate unique identifiers for instances, with different multilingual rendering required.
They have mainly been using F-Logic as their ontology modelling language. This usage was fairly simple in tourism domain, and more sophisticated in the AIFB portal (e.g. use of transitive, symmetric and inverse relations)
Until now, all these ontologies were their own constructions. They have considered using existing ontologies but until now have always opted for constructing their own ontologies.

Ontoprice presented their "time-to-research" portal, which is intended for IT-industry analysts (185 concepts). Particularly interesting about this ontology is that it contains lots of rules, e.g rules for defining the notion of a "competitor" in terms of relations between other concepts.

Document annotation and retrieval

Vaclav Lin from Prague briefly announced their plans for annotating reports with Knowledge and Data Discovery (KDD) results of data-sets. Their intention is to annotate such KDD reports with statistical information in order to enable content-based document retrieval.

Reification

Enrico Motta from KMI/OU-UK described their work on annotation of on-line contents. He argued strongly for reification features as a necessary feature of ontology-languages for such applications. Remarks from the audience suggested that such features are also needed in agent-based applications. Rudiger Klein from Daimler-Chrysler argued that he would need 2nd order constructions (similar to but not the same as reification). Examples where he would need this are:

to qualify properties as being of a certain type (e.g. structural, chemical),
to select different views on same ontology;
for extensibility (e.g to add time-specific constructs to an ontology language if the language itself does not provide them);
to store meta-data in the ontology (e.g source, author, reliability);
even in simple application like annotating abstracts such features would be needed (e.g. to model argumentative structure).

An extensive discussion followed how much of this could be done with the subproperty hierarchy, and how much of this could be done with meta-classes (such as the mechanism provided by Protege). Various people also pointed out that many of these reification and 2nd order constructions are in fact rather simple varieties, and that they need not be cause for computational problems.