RE: data smushing

[Dan Brickley <Daniel.Brickley@bristol.ac.uk>]

On Tue, 2 Jan 2001, David Megginson wrote:

> My #2 was detecting statements that use different identifiers for the
> same entity; there is no reliable way to know whether
> 
>   <rdf:Description rdf:about="http://foo.com/xxx">
>    <dc:title>Joe Smith</dc:title>
>   </rdf:Description>
> 
> and
> 
>   <rdf:Description rdf:about="http://bar.com/yyy">
>    <dc:title>Joe Smith</dc:title>
>   </rdf:Description>
>    
> represent the same entity or not.  Even with the same identifier,
> though, it's hard to be certain whether properties from different
> sources should be treated in the aggregate; imagine that hack.com
> contains
> 
>   <foo:Collection rdf:about="http://foo.com/davidscollection">
>    <foo:item>wooden soprano recorder</foo:item>
>   </foo:Collection>
> 
> and foobar.com contains
> 
>   <foo:Collection rdf:about="http://foo.com/davidscollection">
>    <foo:item>wooden soprano recorder</foo:item>
>   </foo:Collection>
> 
> Does the collection have one or two wooden soprano recorders?  How can
> you be sure?
> 
> In the real world, databases for movies, books, music, etc. tend to
> contain a lot of duplicate entries for authors, etc., and it's hard to
> prune them automatically (I once found over 20 entities for the
> Beatles in a single database).  We'll face the same problem with RDF.

All very true. All I was arguing originally was that in some cases,
further meta-information about properties such as personalMailbox,
corporateHomepage will help us in the aggregation task.

eg I can write something in XML/RDF about Microsoft by identifying the
company indirectly, ie. as "the wordnet:Company with a
util:corporateHomepage of http://www.microsoft.com/". Schema annotations
on the corporateHomepage relation can tell us to only expect at most one
entity for any given value of that property. Better than waiting for
"the" URI scheme for Companies...

I wouldn't claim for a minute that tricks such as this solve the general
data aggregation problem, just that there are some heuristics, hacks and
obvious extensions that go beyond the fiction that we'll have "URIs for
everything".

Dan