[OPEN] and/or [PORT]: Role of Classification


The question of the role of classification/reasoning keeps coming up in various guises, so I suggest a new thread.

As somebody who sees classification as essential, I have strong views on where it is necessary, where it is unnecessary, and where it is inappropriate.

Short message:

A) Authors need classification to author large multiaxial rich ontologies reliably and completely;  Distributed collaborative authorship within a community - the natural mode for many web applications -  particularly need classification to maintain integrity and detect conflicts.

B) Most users will take the outcome of the authoring process including classification, but will not need to do further classification. "Users" in this case include annotators - mechanical or human - of Web resources that use ontology as a source of vocabulary.

C) Hence most ontologies actually published for use will be pre-classified.

D)  Ontologies which are work in progress amongst authors will be at a variety of stages of classifications

E)  An additional consequence of the above is that we need metadata in the ontology to record whether it has been, is intended to be but has not yet been, or is not intended to be classified (and if so with which classifier).

In more detail...

A) When I am creating a classification I want to use classification to manage that process.  I simply do not believe the task of building large multiaxial ontologies consistently is possible without mechanical checking of some sort.  For most use cases, classification is at least an important part of the appropriate mechanical checking.  Even for cases of a mere thousand or so terms, we have good examples of ontologies which were built in weeks with the help of classification which had failed to be built over years without.

B) When I am using an ontology there are two cases;

B1) It is possible to pre-enumerate the concepts that users require..

In this case, once the ontology is classified, the users need know nothing except that they can rely on it. Users can be delivered a pre-classified verified ontology.  This is how we have worked for most applications.  Essentially we use the classifier to help us define a "Pre-coordinated" "phrase book" which is what is used in annotation.   I presume this will be the dominant mode of use in the current phase of the development of the SW.

B2) It is not possible to pre-enumerate the concepts/classes that users require. IE Users potentially require any of a combinatorially large number of concepts that might be defined based on the axioms in the ontology. However, we cannot predict in advance which concepts users will need.  Were we to pre-enumerate all of them, the ontology would be absurdly, perhaps even indefinitely, large.

Some form-generation and data-entry tasks fall into this category.  I have not seen a SW application built this way, but I can well imagine a desire to annotate things richly enough that it would be impossible to pre-enumerate all the concepts required any more than it is possible to pre-enumerate all possible noun phrases in English relevant to a given topic.  In these cases, users need to be given the facility to define new concepts consistent with the ontology and have them classified and placed in it correctly and the other consequences inferred.  (This does not mean that users have to write in OWL or know anything about OWL, but that one output of the user interface would be concepts defined in OWL, which could then be classified, and that classification might influence the further behaviour of the user interface.)  In this case we using the ontology as a "dictionary and a grammar" to drive a "post co-ordinated" application.  How useful such applications will be in the context of the Semantic Web remains to be seen.



Alan L Rector
Professor of Medical Informatics
Department of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL: +44-161-275-6188/6149/7183
FAX: +44-161-275-6236/6204
Room: 2.88a, Kilburn Building
email: rector@cs.man.ac.uk
web: www.cs.man.ac.uk/mig

Received on Monday, 19 April 2004 17:47:48 UTC