"How to Publish Linked Data" vs "Cool URIs don't change"

I wanted to say something about picking URIs during my
lightning talk at an upcoming Semantic Web Dallas meetup.
But I've run into a problem reconciling the recommendations
in HtPLDotW with the recommendations in "Cool URIs don't
change" (and my own experience with enterprise databases
and natural keys)

Specifically, in Section 3 "Choosing URIs" of HtPLDotW,
the suggestion to use mnemonic names (names linked to
some of the data about the object) has proven to be
problematic in practice. The "What to leave out" section
of "Cool URIs don't change" is very explicit that this is an
anti-pattern and should be avoided.

Now, there's a legitimate balancing act between using the
"don't depend on natural keys" best practice and minting
Linked Data URLs that are amenable to pattern-based links,
but for the most part doing things like including the word
"Berlin" in a canonical URI is just asking for trouble in the
long (200 year?) term (just ask the residents of Bombay)

There's an escape hatch if you choose to use a widely
recognized naming scheme (like, say, ISBN numbers), but
even then there are risks if the semantics of the identifier
scheme don't match _exactly_ with the way the scheme is
being re-used (ISBN numbers don't correspond to books in
the way most people expect, and the fact that they can be
re-used generally comes as a surprise)

But, before I go saying critical things about HtPLDotW in
front of a live (and well informed) audience, I wanted to
do a sanity check.

Thanks for any feedback,

-cks

-- 
Christopher St. John
cks@praxisbridge.com
http://praxisbridge.com
http://artofsystems.blogspot.com

Received on Friday, 10 July 2009 05:43:37 UTC