Re: [SKOS] The return of ISSUE-44 (was Re: TR : SKOS Reference Editor's Draft 23 December 2007)

At 11:07 AM 1/10/2008, Bernard Vatant wrote:
>OK Daniel, let me have another try  if you don't mind  :-)
>>
>> From my point of view, it does NOT make sense that skos:narrower 
>> and broader are not transitive.
>>And if applications can go ahead and make them transitive by 
>>expanding how they wish, that violates the asserted SKOS semantics. 
>>Unless I'm misunderstanding something here, this sounds like a 
>>formula for chaos.
>Expanding the query does not *make* the relation transitive, it's 
>just an application feature. I don't see any violation of the 
>semantics. The results proposed are not results of the original 
>query, but from *query expansion*. The query expansion is not the 
>original query, right? There is one single way to strictly answer 
>the query, and many ways to expand it.
>
>I have this real-life example at a customer's in legal publication.
>The figures are around 2 million documents, and 50,000 concepts in 
>the vocabulary (and growing), with a very deep tree.
>Suppose I start a search at level 3, on a concept with 5 direct 
>narrower concepts, and about 500 more downwards if transitivity is applied.
>If I don't expand the query, say I get 40 answers indexed on the 
>direct narrower concepts, if I expand it with unbound transitivity, 
>say I get 4,000 answers. Way too many. Think about performance.

My point is that the relation semantics should determine whether it 
is appropriate to expand a term transitively. I got the sense from 
prior discussion that people were talking about doing that regardless 
of transitivity.

> From a end-user perspective, what is the best? Retrieving very 
> quickly the 40 resources classified directly by the 5 direct 
> children, and allowing the user to expand from one of those one or 
> two steps down, does not seem a recipe for chaos, but for a sound 
> adaptation to the context, and for tackling some scalability 
> issues. If transitivity is built in the semantics, I have to go 
> down the tree and retrieve the 4,000 answers. If I want to trim the 
> tree to limit the results, there I will break the semantics ...
>
>Does that make sense?

How many levels an application decides to expand is independent of 
whether it is semantically correct to expand along that relation in 
the first place.


>Best
>
>Bernard
>
>
><http://mondeca.wordpress.com/>
>

Received on Saturday, 12 January 2008 13:58:43 UTC