My purpose is to briefly highlight a number of points for discussion raised by our work done with ontologies. This work has been done re-engineering business information systems, primarily in the financial sector. So business objects are, among other things, financial products (see for example: Partridge, Chris, Business objects: re-engineering and re-use, Reissue 1995. 250 pp. paperback, Butterworth Heineman, ISBN 075062082X).
We found that an important starting point for ontological modelling is the core top ontology. The one we have developed is shown in the slides. We found that it was important to choose a sufficiently flexible (expressive) top formal ontology. If this was inflexible, we found it became difficult to subsume properly when generalising. We based our formal ontology around the set-theoretic concepts of class and tuple. From a practical point of view, we found these really effective tools for both building a coherent ontology and generalising.
At the formal level, we use two key relations;
There is a tendency in some ontologies to unnecessarily constrain these relations, assuming that it makes things simpler. The tendency to constrain the super-sub-class relation restricts each sub class to one and only one super class. There is also a tendency to constrain the core class-member relation, allowing each member to only belong to one and only one class. This is known as a single classification structure. While it may make the individual relations look simpler, it makes the overall model much less simple.
These constraints severely hinder generalisation and subsumption. Subsumption is where a combination of more general objects subsumes all the details of a combination of less general objects. So, from one perspective, the less general objects are superfluous and can be eliminated from the model. Subsuming generalisation follows a general pattern (shown in the slides). It is one of the most common patterns in our re-engineering work.
The benefits of subsuming generalisation are usually more than just compressing more knowledge into fewer, more general classes. Typically the general classes are also fruitful; they handle much more than the details from which they have been generalised. Part of the reason for the fruitfulness is that generalisation does more than just simplify, it also clarifies and fills out patterns.
Below the formal level is the spatio-temporal level. This has as its
key relation whole-part tuples.
At this level, an ontology has to decide on what the particulars (the 'things'
in this model) are. Are they three-dimensional, four-dimensional or something
else? We found that it was useful to think in terms of four-dimensional
things, where time is the fourth dimension. This four-dimensional ontology
greatly simplifies the handling of identity over time.
In summary, our experience with ontologies was that it was important to start with a good core top ontology. It was particularly important to start with a sufficiently flexible (expressive) top formal ontology element.
ProKSI-97
Report