Vladimir,

I'm thinking of trying to do some stats on the existing ontology and the 
mappings to see where there is room for improvement.  I'm tied up this week 
with a couple deadlines that I seem to moving towards at greater than light 
speed, though my progress is not.

As soon as I get the rough cut done, I'll share the results with you and maybe 
we can discuss paths forward?

I'm with you on the 30% error rate...that doesn't help anyone.

Aaron

On Feb 25, 2015, at 08:02, Vladimir Alexiev <[email protected]> 
wrote:

>> From: M. Aaron Bossert [mailto:[email protected]]
>> I am more than happy to work the ML problem with you.  
> 
> Hi Aaron!
> Would be great to work with someone from Cray but I don't have a good idea 
> how to use ML here,
> nor indeed a lot of trust in using ML to produce or fix mappings.
> 
> E.g. see this exchange:
> https://twitter.com/valexiev1/status/565814870973890560
> Generating 30% wrong prop maps for the Ukrainian dbpedia is IMHO doing them a 
> disservice!
> Who's gonna clean up all this?
> 
> I guess I'm more of a MLab (Manual Labor) guy, I just learned they coined 
> such alias for crowdsourcing:
> http://link.springer.com/chapter/10.1007%2F978-3-319-13704-9_14
> 
>>> DBO: dbo:parent rdfs:range dbo:Person
>>> Wikipedia: | mother = [[Queen Victoria]] of [[England]]
>> For your example of the dichotomy with the domain and range of "mother" and
>> queen Victoria being the "mother", this begs for contextual approach to that 
>> concept....
> 
> She IS the mother, not sure what you mean.
> 
> Here a simple post-extraction cleanup can take care of it: 
> remove all statements that violate range (so dbo:parent [[England]] will be 
> removed).
> But we dare not do it, because many of the ranges are imprecise, or set 
> wishfully without regard to existing data / mappings.
> (As usual, the real data is more complex than any model of it.)
> 
> So we need to check our Ontological Assumptions and precise domains/ranges 
> before such cleanup.
> See example in 
> http://vladimiralexiev.github.io/pres/20150209-dbpedia/dbpedia-problems-long.html#sec-6-7
> 
> 

Reply via email to