Re: HL7 RIM Designtime OWL Runtime RDF

I am enjoying reading and catching up on this thread. 

David, you mentioned 'rdf model' below - are you referring to ontology models? 

And, you said "To my mind, monotonicity is the key."  But in medicine most reasoning is non-monotonic  - default reasoning, (educated) guesses and revision of diagnosis as new data comes into the picture. What am I missing here?

thanks,
Sivaram
____________________________
Sivaram Arabandi, MD, MS
ONTOPRO
www.ontopro.com
Ph:  832.726.2322

http://ontolog.cim3.net/cgi-bin/wiki.pl?SivaramArabandi
http://www.linkedin.com/pub/sivaram-arabandi/1/9ab/92a





On Jan 16, 2013, at 1:46 PM, David Booth wrote:

> Hi Peter,
> 
> On Wed, 2013-01-16 at 08:39 -0800, Peter.Hendler@kp.org wrote:
>> Eric et al, 
>> Is there any material on the idea of "design time OWL runtime RDF"?   
>> 
>> Is it Kosher, once you are done with your reasoners, to convert to RDF
>> and then treat it as if it were closed world like a database? 
> 
> Absolutely.  Almost all applications use a closed world assumption at
> some point.
>> 
>> RIM which is OO, is of course closed world and can be represented in a
>> database. Nothing can change, no new assertions can be made. When an
>> HL7 message is sent, we assume it can't be changed by a reasoner or
>> anything else. It is set in stone. In fact, there are laws. You are
>> not allowed to edit a message once it's been sent. 
>> 
>> In open world, anyone can add to the triple store at any time, and
>> meanings can change. But in an HL7 message, once you make the message,
>> you are not allowed to amend or add to it. 
> 
> It should be monotonic, so even though all of the existing statements
> still hold, additional statements may be true also.  When talking about
> "changing" some data, it's important to distinguish between
> (monotonically) adding more data to it and (non-monotonically) modifying
> the existing data.
>> 
>> On a related note.  We have different ways of expressing negation to
>> Acts. Much of the complication comes from whether the negation is done
>> in the vocabulary (SNOMED) or the OO part of the model (RIM). 
>> How can we tell if two different representations where the is negation
>> expressed on different parts in the model, are semantically the same? 
> 
> Can you give an example? 
>> 
>> The terminology (SNOMED open world, OK to use reasoners) and the RIM
>> (OO closed world) can not be mixed (I think).  
> 
> Why do you say that they cannot be mixed?  You do have to be careful to
> know which data is making what closed world assumption.
> 
>> But my question is this. 
>> Can you reduce the whole representation of the RIM part of the model
>> and the terminology part (SNOMED) into one set of triples, and then
>> could you reduce two  instances of the these mixed models to graphs of
>> triplets that you can compare? 
> 
> In principle, yes, I think so.  But let me turn it around the other way.
> I think it is important to design the RDF models such that they can be
> mixed and instances can be compared.  If there are problems in doing so,
> then we need to correct the models to fix them.
>> 
>> If you did reduce/normalize the mixed model of the OO RIM and the EL+
>> logic SNOMED into one set of triples.  Could you consider these, for
>> your comparisons, as if they are closed world and simply compare the
>> graph patterns? 
>> 
>> This is another way to ask.  At any point in the life of a model (HL7
>> message or clinical statement for example), can you just declare "from
>> this point forward, no one is allowed to add to or change this graph
>> in any way", and then treat the whole graph as if it is closed world,
>> even though at an earlier point in the graphs life cycle it did
>> consist of SNOMED (open) and RIM (closed)?  Does it become closed by
>> agreement not to add to it after it is final? 
>> 
> I think it is critical that the RDF models be designed to be monotonic,
> so that you can always add more information without invalidating
> previous information.  This means that you cannot just say something
> like "Mary is pregnant".  It has to be qualified to a particular context
> or time period, such as "On 1-Jan-2013 Mary's pregnancy test was
> negative".  (Sorry for such an obvious example, but hopefully you see
> what I mean.)
> 
> To my mind, monotonicity is the key.  I normally think of "closed world"
> and "open world" as being more about what you *do* with the data, than
> being about the data itself.  If the data is designed to be monotonic,
> then for specific uses you can use closed world reasoning.
> 
> With all that said, I'm not certain that I'm really hitting on the
> question that you're raising, so if you can show a more concrete example
> it may help.
> 
> Thanks!
> 
> -- 
> David Booth, Ph.D.
> http://dbooth.org/
> 
> Loss of web prodigy Aaron Swartz: http://tinyurl.com/ahe2k8c
> 
> Opinions expressed herein are those of the author and do not necessarily
> reflect those of his employer.
> 
>

Received on Wednesday, 16 January 2013 20:51:57 UTC