O)K... so Phil came back and commented on my post from yesterday....

One suggestion was to support mapping between tags. I still have issues with ontological mapping like this as I think that it remains both error prone and a time consuming PITA. But, I did some thinking while abluting this morning (the little boys room is a great cognative stimulant) and I reckon that we must be getting close to being able to draw inference mappings from real world data.

What do I mean from this. Well. Google has a whole schwag of information on various sites around the world that would, IMHO, cover pretty much every conceivable topic. One would think that they should be able to use their Googlebase to build some sort of inference engine that would give a ranked inference between potential tag terms. So, from my example yesterday, Canoeing and Kayaking would sit very close together on the inference list. Thus tag ontologies and the mappings between them could become a fuzzy concept and we could use fuzzy algebra to help us build systems around them.

I played around with this sort of autmatic inferencing in music files in my final year AI paper @ varsity. Used Self Organising Maps to cluster similar music. Google could use some sort of Neural Net to cluster key words out of their database. Then this cluster map could be used to help map between concepts in a fuzzy fashion.

The advantage of this approach is that these maps could then be represented to the user in 2D or 3D space and coupled with some funky zooming could be used to help solve the UI problem that Phil talks about.