W3C home > Mailing lists > Public > semantic-web@w3.org > June 2012

Re: Reasoner accuracy

From: Ghalem Ouadjed (EOWEO) <gouadjed@eoweo.com>
Date: Sat, 23 Jun 2012 10:37:26 +0200
Message-ID: <4FE58046.5020706@eoweo.com>
To: adasal <adam.saltiel@gmail.com>
CC: "semantic-web@w3.org" <semantic-web@w3.org>
Le 22/06/2012 17:40, adasal a écrit :
> Well it's a very complex subject isn't it.
> I have never done reasoner optimization but e.g. Allegro claim their 
> reasoner is faster over a certain data set than some other X. And I 
> think theirs and others have reasoners which are plug in. So the first 
> step is understanding the significance of the underlying data store.
> Then there is the logic the reasoner supports. Some are opptimised for 
> different branches. But may do less well than X with some other logic set.
> I think choice of logic comes before choice of reasoner though?
> So now we have the store, the logic, the reasoner and add in the 
> implementation language and the query language.
> If it is a complex store (Open RDF?) we may also be looking at its 
> component modules and their implementation.
> Don't forget versions.
> Now what do you want to know?
> (not just to be clear that I would be able to answer. But then think 
> about it very few people would given above. )
>
> Adam
Hi Adam,

yes the first thought when the users talk about their prob using 
reasoner on their data concerns the data. Actually they all use Pellet 
and i can add that their data is not so clean because they are produced 
from other semi structured data (xml like). As for today the users have 
a conservative process which consists on preserving their initial format 
and produce some rdf/xml files in a way to enhance the conclusion. But 
these enhancements are not stable. For example the conclusions they get 
are " not always equivalent". if we consider that the data is 80% 
responsible what would the format the most interesting ? My thought is 
that turtle like is interesting as for me the reasoners we use to meet 
are Prolog based (thought?) and then N3 + rules could provide better 
results...?..

Ghalem
Received on Saturday, 23 June 2012 08:37:58 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 21:45:49 GMT