[Sigia-l] Findability - hierarchies

[Cunliffe D J (Comp)]

Greetings,

> From: Lars Marius Garshol [mailto:larsga at garshol.priv.no]
> Faceted metadata essentially means replacing a single hierarchy by
> multiple hierarchy, with one hierarchy for each facet. Now,
> structurally that's actually the same as having a single hierarchy.
> (Just add a single node at the top linking all the facets and you've
> got it.)

This is of course true, but we can trivially have a conceptually equivalent
effect by adding an 'everything' category to whatever classification scheme
we choose. It does seem to misrepresent the purpose of facets however, which
is to reflect fundamentally different aspects of the things we are
categorising (such as business areas and geographical locations).

> And hierarchies fail on several counts in this regard:

Many of these counts seem to impose a very restrictive view of hierarchies
(or at least thesauri) in practice (as opposed to a pure theoretical
viewpoint)

>  - you cannot type relationships, so there is only a single kind of
>    relationship between the nodes,

Many thesauri have typed relationships beyond the BT/NT hierarchical
relationships. Often this typing is only weakly typed (in the programming
sense) but there is a lot of interest in more strongly typed relationships.

>  - you cannot type the nodes, so a machine can't tell countries from
>    diseases from people from animal species,

But this is one of the things you get from facets - diseases and countries
might well be modelled and categorized as fundamentally different things.

>  - you cannot assign properties to the nodes beyond one or two fixed
>    kinds of names and perhaps some untyped URIs to content, and

I'm not quite sure what you are meaning here - what type of properties might
we want to store?

>  - your relationships must form a tree.

Even the BT/NT hierarchy at the centre of the taxonomy may sometimes be a
polyhierarchy (more than one parent) rather than a strict hierarchy. The
introduction of other relationship types typically cuts across the hierarchy
or across facets.

> I think you misunderstood what I meant by automation: I was thinking
> of having a machine use the categorization system. Of course, so long
> as you are talking about simple categories that won't work, but with
> an ontology the machine can do more than show you a tree. (See below
> for a trivial example.)

Even a strict hierarchy allows some limited automated reasoning, such as
retrieval by semantic similarity.

> What you'd really want to say is something like this:
> 
>   "Oil services" is a "business area"
>   "Oil surveying" is a "department"
>   "Kano field location" is an "office"
>   "Oil production" is a "department"
> 
>   "Africa" is a "continent"
>   "Nigeria" is a "country"
>   "Kano" is a "place"
>   "Morocco" is a "country"
> 
>   "Kano field location" is "located in" "Kano"

Which looks to me like a faceted hierarchy with a specialised type of
related-to relationship representing the 'located in' concept.

Maybe we are just arguing semantics again. An interesting question is where
we draw the line (if indeed we do) between a thesaurus and an ontology. Do
we need to go as far as Doug Lenat's Cyc or can we develop more pragmatic
solutions even if they only apply to specialised domains?

Be seeing you,

Daniel.