- From: Antoine Isaac <aisaac@few.vu.nl>
- Date: Mon, 10 Dec 2007 21:26:05 +0100
- To: Stella Dextre Clarke <sdclarke@lukehouse.demon.co.uk>
- CC: 'Leonard Will' <L.Will@willpowerinfo.co.uk>, public-swd-wg@w3.org, public-esw-thes@w3.org
Hi Stella, >> The only situation where I can think of problem is when applications >> deal with purely extension-based links that have no clear >> thesaurus-like >> semantic link, like the "France"/"war" I used in [2]. But >> even if this >> kind of link cannot be made relatedMatch (which can also be >> argued, in >> fact) we can hypothesize that these would be very specific >> applications. >> And we would encourage them to define their own >> relationships. Which by >> the way they are likely to do, if they focus only on >> documents, e.g. by >> introducing different levels of extensional overlap (30%, 50%, etc) >> > Relationships defined on the basis of a particular document set or > collection or pair of collections have some very useful applications, it > seems to me, but these are different from what I imagine SKOS is > generally for. If a relationship is a "property" of pairs/groups of > concepts within a scheme, and a mapping is the same type of thing > applying to pairs/groups of concepts where the concepts come from more > than one scheme, then a match that takes into account the occurrence in > a document collection is a property of this system as a whole, that is > to say, it is a property relating to > one-or-more-schemes-plus-one-or-more-document-collections. So the > document collection(s) to which it is being applied need to be included > in the scope statement of the data being communicated. And that makes it > all more complicated. You are right! > Furthermore, the applications where statistical > occurrence data are useful tend to be free text applications, whereas I > think of SKOS concepts as useful mostly in the context of indexed > collections. Sorry, no time to express this carefully and rigorously, > but I hope the general drift of the argument makes sense. > Yes, it does. Even though I have been involved in experiments showing that (at least in some contexts) extensional information brings some potentially useful similarity information [1] Best, Antoine [1] http://www.few.vu.nl/~aisaac/papers/STITCH-Instance-ISWC07.pdf
Received on Monday, 10 December 2007 20:41:57 UTC