W3C home > Mailing lists > Public > public-lod@w3.org > April 2010

Re: [Patterns] Materialize Inferences (was Re: Triple materialization at publisher level)

From: Leigh Dodds <leigh.dodds@talis.com>
Date: Sat, 10 Apr 2010 11:15:06 +0100
Message-ID: <t2pf323a4471004100315rfeac60bs69fd01a7024029c5@mail.gmail.com>
To: Patrick Logan <patrickdlogan@gmail.com>
Cc: public-lod <public-lod@w3.org>

On 9 April 2010 17:59, Patrick Logan <patrickdlogan@gmail.com> wrote:
> As I understand it, the most concise approach would be to use the
> pattern "Equivalence Links". Then to make those more maintainable,
> possibly use "Link Base".

For sameAs links yes that's an approach. In fact you've just
highlighted that there's a more basic pattern: publishing key data
with seeAlso to additional data. I've described those as Enhanced
Descriptions [1].

So the general form of that underlying pattern is that additional data
is factored out into separate documents, referenced by See Also links.
Those documents contain Annotations of the original resources.

One realisation of that pattern is a Link Base. Another is to
separately Materialize Inferences. A third, as I described in my blog
post, is to separately publish access controlled "premium" data.

> But the "Materialize Inferences" is indicating there are forces on the
> data provider to perform those inferences over equivalent links at the
> source, and to make (materialize) the resulting links explicitly?
> What are those forces that would lean a developer one way or the
> other? They seem to be based on the capabilities of the data
> _consumer_.
> Do you (always?) provide two options? "Give me the concise equivalence
> links" and "Give me all the materialized inferences?"

As Vasiliy mentions the forces are dependent on the potential
consumers of the data being published. Those forces might vary
considerably within an enterprise, across a B2B exchange, or on the
open internet. A publisher opting to use the pattern (or needing to
because of specific client requirements) will then need to decide how
best to deploy it.

I'd suggest that some inferences are relatively "harmless" if included
directly (e.g. label variants) but others, e.g. that infer more
structure in the data may always be better to publish separately to
allow clients more choice.



[1]. http://www.ldodds.com/blog/2010/03/enhanced-descriptions-premium-linked-data/

Leigh Dodds
Programme Manager, Talis Platform
Received on Saturday, 10 April 2010 10:15:38 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 15:16:05 UTC