Bottom-up ontologies

Bottom-up ontologies

Ontologies are artifacts used to reduce gaps in the interpretation of data between peers.

The literature commonly defines an ontology as adopted agreements about an observed reality within a particular context, which is represented in a way so machines can process it.

It consists of a finite set of agreements, which in turn is abstracted into primitive representations, constituting the necessary inventory to define new concept representations via composition. This set of primitives, also called upper ontology, aims to provide interchanging information between peers, via representations such as RDF that can be adequately interpreted by peers within a defined context or domain.

Ontologies traditionally were constructed by committees or organized groups that act as “authorities” on a specific domain are especially useful in contexts where we can perceive unambiguous “truths” within a context, such as describing the price of a good, the wavelength of a specific color, the current weather and so on.

But when we model subjective appreciations, the scenario is quite different, since the perceived truth is diffuse, tending to ambiguous terminology, and categorizations. It is more convenient to count with multiple and simultaneous “ontologies,” which are mapped together in those cases. These mappings should be built incrementally by all the relevant parties because a central authority cannot necessarily interpret all the ontologies correctly.

This last approach is called ‘bottom-up.’ Unfortunately, it hasn’t been explored extensively through the literature.

There is no definitive metamodel to suit all needs. Usually, organizations make use of different metamodels at once. How to mix effectively separate metamodels remains an open question usually answered by observing the peers’ interaction.

A bottom-up approach

To build ontologies in a bottom-up approach one have to put special attention to how the peers interact within a context. How they communicate and exchange views about subjects, how they incrementally agree on how to enable information exchange. Possibly having schema hubs, where users can clone and modify schemas as Github does with code. Ontologies can be decomposed, parts of them can evolve indefinitely and others will be stable enough to form part of upper-ontologies that are valid for all communities.

There is no such thing as the definitive model.