- From: Charles McCathieNevile <charles@sidar.org>
- Date: Mon, 18 Apr 2005 19:47:44 +0200
- To: "Giorgio Brajnik" <giorgio@dimi.uniud.it>, public-wai-ert@w3.org
On Mon, 18 Apr 2005 16:33:33 +0200, Giorgio Brajnik <giorgio@dimi.uniud.it> wrote: > I would suggest to consider confidence factors as probabilities > associated to assertions (like "this test has failed"). Yes, this is how i understand the confidence information, and I think what the general understanding is in the group. The quetion is how to describe it. Some tests will have fairly clear and detailed confidence parameters, others won't. RDF allows us to say which is which, in our results, if we model the information correctly. Which means that when we are trying to compare confidence across different kinds of test results for the same problem, we can be clear about what we are comparing. If we decide on one confidence scheme then we have to try to push everything to fit it, which doesn't seem sensible. If we have multiple ones, then for any particular use case we need to define some relation between them, but we can still know in the results that they are different ones, and change the mappings at that time rather than lose the fact that they are different at the point where we record the results. For example, if Chris uses "high, medium, low" and Giorgio uses a number from 1 to 7 and Nils uses an integer from 0 to 100, I can map Chris' confidence to 1, 4 and 7 on Giorgio's scale and map that to some numbers from Nils. Equally, I can decide to do a little more work and map some of Chris' results to 1, 4 and 7, some of them to 3, 5, 7 and some to 1, 3, 4 according to what the test is... Does this sound like what others are thinking? cheers Chaals -- Charles McCathieNevile Fundacion Sidar charles@sidar.org +61 409 134 136 http://www.sidar.org
Received on Monday, 18 April 2005 17:47:57 UTC