Spatial identification of subjects

Hi,

First of all, thanks to the organisers of TMRA for an absorbing 
conference, and a chance to re-acquaint myself with the Topic Maps 
scene.

One talk which got me thinking was Sven Krosse on identity for places. 
I'm afraid I don't agree with this approach, and wonder what others 
think.

As it happens, I just went across to Subj3ct.com and found that there is 
a steady stream of these "geo:" identifiers arriving there.  A random 
example is:

geo:51.7,20.5333333;crs=wgs84

This is equivalent to the Geonames concept 774834:

http://sws.geonames.org/774834/about.rdf

The first problem I have with these identifiers is that they are not 
dereferenceable.  You can clearly see this if you click on the link 
helpfully provided on the Subj3ct page for this identifier.  ("Firefox 
doesn't know how to open this address, because the protocol (geo) isn't 
associated with any program.")  Therefore they have no value as Linked 
Data identifiers.

The second problem is that it is far from clear to me how users are 
meant to find out the exact coordinates of the centroid of a place, and 
use those to generate the "right" identifier for that place. Bujaly is a 
small place, but even there I see that there are already two distinct 
"geo" identifiers associated with it, even before any users come along 
and try to guess what the "right" coordinates might be.  How much harder 
would it be for a place the size of London?

The third problem is one of semantic overloading. A WGS84 coordinate is 
just that, and should in the Linked Data scheme of things have its own 
identity.  Dropping onto that identity the additional load of 
representing a populated place simply confuses the issue.  Also, there 
are different types of "place": popular usage; administrative areas; 
political constituency boundaries; etc.  What happens when two such 
entities happen to have the same boundaries and therefore the same 
centroid?

It is clearly tempting to look for a "magic bullet hash function" which 
will generate shared unique identifiers, but I don't think that this is 
the right thing to do.  In general, I would go for a strategy where your 
identifiers are deliberately lacking in significance, like Geonames' 
numerical identities, and you use the properties of a published subject 
to determine equivalence.

In this case, the original Geonames RDF gives you the place name and the 
WGS84 coordinates.  It would be a simple matter to match on the name, 
and then check that the coordinates are "sufficiently close" to your own 
coordinates for that place, to be sure that the Geonames identifier 
refers to the same place.  Another strategy, which would be more 
relevant for the sort of museum data I am familiar with, would be to use 
the containment relationships between this place and the larger 
geopolitical units it falls within.

Clearly, deriving and sharing unique persistent identifiers for 
subjects/concepts is a major challenge.  I think this challenge is best 
met by publishing as many (machine-processible) properties with a 
published subject identifier as you can, so that others can deduce 
(possibly algorithmically) whether the entity they are considering 
matches the one you have published.  We may, for example, end up with 
"virtual merges" [1] which have a numerical degree of confidence 
associated with them.

Richard

[1] Jack Park, Topic Map scenarios for Knowledge Federation, TMRA 2010

-- 
Richard Light