Re: Can we lower the LD entry cost please (part 1)?

On 07/02/2009 19:57, "Yves Raimond" <yves.raimond@gmail.com> wrote:

> Hello!
>
>> Sorry, I just cannot accept that a SPARQL endpoint is th esort of thing that
>> we should be expecting new casual users to try to use, even with a query
>> builder.
>
> You made the point about "linkage systems" - I was answering to that.
Ah, OK - sorry to misunderstand.
And in fact my comments below were not intended to address linkage systems
either.
> I am not suggesting casual users should write SPARQL queries.
>
>> And I am not even sure I accept the "more reliable" argument.
>> When I search for a string I get back a resolvable URI, which leads me to
>> exactly the information I need to know to decide if I have the right one (or
>> it should).
>
> No! You have *absolutely* no way to guess how the results were
> constructed! For example, the results will be completely different if
I agree - but that is true if I ask a SPARQL endpoint what it knows about
the same URI.
But as I said, I was not talking about the situation of building a search
engine, or linkage system.

Best
Hugh
> you built your search engine on top of lucene index of all rdfs:label,
> of all rdfs:comment, of all literals, of all literals of neighboring
> resources, etc. If you use these results for automated interlinking,
> you're putting quite a lot of trust in the search engine provided. On
> a side-note, we mentioned similar problems in our LDOW paper from last
> year. Incorporating a "black box" in an automated interlinking
> algorithm (in our case, MusicDNS fingerprinting) is really, really
> tricky.
>
>
>> And even for a SPARQL query you have no more idea how I built the knowledge
>> than a resolvable URI.
>
> You clearly have more information, as you end up constructing your
> similarity measure yourself (SPARQL doesn't do fuzzy matching).
>
>
> Cheers!
> y
>

Received on Saturday, 7 February 2009 21:18:38 UTC