Hi Robert, it looks to me that you do not need any fancy features to derive the intended inferences - since you have two (mutually exclusive) kind of facts that will be present there is no need to make things overcomplicated.
Let me give you a simplified example how you could achieve that using the
Examination/Symptom/Diagnosis domain (the example rdf is attached):
1. Lets define a class ex:Symptom which and populate it with few instances:
ex:SymptomA, ex:SymptomB and ex:SymptomC
2. Then we could model the presence/absence of some symptom using two
additional classes: ex:SymptomPresent and ex:SymptomNotPresent (and optionally
define those disjoint)
3. The main idea is to use two (mutually exclusive) properties to state a
particular presence or absence of some symptom during a medical examination,
lets name those propeties: ex:haveSymptom and ex:doesNothaveSymptom
4. for each of the Symptom instances, we define a pair of subclasses of,
repsectively, ex:SymptomPresent and ex:SymptomNotPresent which are modelled as
owl:hasValue owl:Restrictions over the above two properties, e.g. for
ex:SymptomA, those look like:
ex:HaveSymptomA rdfs:subClassOf ex:SymptomPresent ;
rdf:type owl:Restriction ;
owl:onProperty ex:haveSymptom ;
owl:hasValue ex:SymptomA .
ex:DoesNOTHaveSymptomA rdfs:subClassOf ex:SymptomNotPresent ;
rdf:type owl:Restriction ;
owl:onProperty ex:doesNothaveSymptom ;
owl:hasValue ex:SymptomA .
that way, when you assert a ex:haveSymptom or ex:doesNothaveSymptom relation
between something (Examination instance) and a particular Symptom, then that
Examination will be classified as instance of some of the above Restrictions.
5. Then you could model a combination of presence/absence of Symptoms as
instances of some top ex:Diagnosis class e.g.
ex:Diagnosis rdf:type rdfs:Class .
ex:Diagnosis_haveAnotB rdf:type ex:Diagnosis ;
owl:intersectionOf (
ex:HaveSymptomA
ex:DoesNOTHaveSymptomB
) .
ex:Diagnosis_haveCnotA rdf:type ex:Diagnosis ;
owl:intersectionOf (
ex:HaveSymptomC
ex:DoesNOTHaveSymptomA
) .
.... and so on ...
Having such model of the domani, then an instance of a medical examination
could be used to group together the data about the person examined and the
presence / absence of symptoms, which due to the inference, will be classified
as instance of a particular Diagnosis, as defined above. If the right relations
are asserted, e.g (like these below.:
ex:Person rdf:type rdfs:Class .
ex:John rdf:type ex:Person .
ex:Marry rdf:type ex:Person .
ex:Examination rdf:type rdfs:Class .
[] rdf:type ex:Examination ;
ex:about ex:John ;
ex:haveSymptom ex:SymptomA ;
ex:doesNothaveSymptom ex:SymptomB .
[] rdf:type ex:Examination ;
ex:about ex:Marry ;
ex:doesNothaveSymptom ex:SymptomA ;
ex:haveSymptom ex:SymptomC .
You could query about the classified diagnosis using a query that may look like:
prefix ex: <http://example.org/example#>
prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#>
prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
select * where {
?ex a ex:Examination .
?ex ex:about ?person.
?ex a ?diag .
?diag rdf:type ex:Diagnosis .
}
which gave me the following resultset when I put the attached RDF into a
repository preset with "owl-horst-optimized" ruleset:
ex person diag
_:node66 ex:John ex:Diagnosis_haveAnotB
_:node67 ex:Marry ex:Diagnosis_haveCnotA
Of course, using that approach, you may end with multiple diagnoses for an
examination in case some diagnosis is a complete subset of another.
Hope that this is of some help to you.
Best,
Damyan Ognyanov
Ontotext AD.
----- Original Message -----
From: Buchmann Robert
To: owlim-discussion@ontotext.com
Sent: Tuesday, February 01, 2011 9:24 PM
Subject: Re: [Owlim-discussion] OWLIM-discussion Digest, Vol 25, Issue 4
Hi,
From what i understand, the open world assumption means that you cannot
infer/draw conclusions from the ABSENCE of triples (because they could be added
in the future).
In my interpretation this does not mean that you shouldn't draw conclusions
from NEGATED triples.
Maybe i was vague about what i call "negated triples". I think it's more
precise to call them FALSE triples: its when a relationship does NOT hold
between a subject and an object and i want to explicitly state that, then infer
based on that statement. Something like:
:TheSun :DoesntOrbitAround :TheEarth
(i guess i can express falseness like this...and state that :DoesntOrbit is
disjointPropertywith :DoesOrbit? it just popped in my mind right now)
However it's a bit too naming-convention-based for my taste. I thought that
it would be more elegant to assign a context of :False in Sesame:
:TheSun :OrbitsAround :TheEarth (context :False)
Since SPARQL allows for filtering based on named graphs (=contexts) it would
be very easy to separate the False triples from the True (as long as there's no
inference involved).
Now coming back to my problem - i'd appreciate a lot an advice about how to
model the rule i was talking about:
If X has symptom A, but NOT symptom b => X has diagnosis D.
(again, not that we dont know that x has symptom b, but rather that we
explicitly know that x does not have that attribute).
Thanks,
Robi
------------------------------------------------------------------------------
From: "owlim-discussion-requ...@ontotext.com"
<owlim-discussion-requ...@ontotext.com>
To: owlim-discussion@ontotext.com
Sent: Tue, February 1, 2011 7:00:03 PM
Subject: OWLIM-discussion Digest, Vol 25, Issue 4
Send OWLIM-discussion mailing list submissions to
owlim-discussion@ontotext.com
To subscribe or unsubscribe via the World Wide Web, visit
http://ontotext.com/mailman/listinfo/owlim-discussion
or, via email, send a message with subject or body 'help' to
owlim-discussion-requ...@ontotext.com
You can reach the person managing the list at
owlim-discussion-ow...@ontotext.com
When replying, please edit your Subject line so it is more specific
than "Re: Contents of OWLIM-discussion digest..."
Today's Topics:
1. Re: newbie questions - the sequel: contextual inferencing and
mutual exclusive properties (Ivan Peikov)
----------------------------------------------------------------------
Message: 1
Date: Tue, 1 Feb 2011 17:27:01 +0200
From: Ivan Peikov <ivan.pei...@ontotext.com>
Subject: Re: [Owlim-discussion] newbie questions - the sequel:
contextual inferencing and mutual exclusive properties
To: owlim-discussion@ontotext.com
Cc: Buchmann Robert <buchma...@yahoo.com>
Message-ID: <201102011727.01931.ivan.pei...@ontotext.com>
Content-Type: Text/Plain; charset="iso-8859-6"
Hi Robi,
Here are my thoughts on your questions:
1. Normally, in monotonic rule systems that adhere to the open world
assumption (such a OWLIM and the semantic web in general) one doesn't use
negations in the rules' premises because knowledge can be added at any time
and thus trigger the retraction of previously inferred knowledge from the
system. Therefore, allowing negation in the rule language would make the
system non-monotonic.
Even though you are not allowed to express negative assertions in the rules,
however, you can achieve the desired behavior query time by forcing the
negative constraints in e.g. a SPARQL query.
As for the context of the inferred triples - indeed OWLIM puts all inferred
triples into the so called "implicit" (or also "anonymous") context. There is
no system setting that can change that behavior but once again, I believe you
don't really need to.
2. You are looking for owl:propertyDisjointWith. It is implemented in OWL2-RL
and OWL2-QL profiles in OWLIM.
Hope this helps!
Cheers,
Ivan
On Tuesday 01 February 2011 12:39:04 Buchmann Robert wrote:
> Hi,
>
> 1. Context inferencing.
>
> I noticed that under Sesame (including OWLIM) all inferred triples are
> assigned to a new, anonymous context, other than the ones that i assign at
> upload time. In other words, you can upload as many named graphs as i
> want, but i cannot filter inferred triples based on it.
>
> I came to this while trying to define some custom rules in the PIE files,
> in order to simulate a medical differential diagnosis.
> Something like: If you have symptom A, but NOT symptom b => diagnosis D.
> Now my problem is how to express negation in a custom rule. Actually im not
> sure how to express it in RDF at all.
>
> I was thinking about using the context as a truth value:
> {:John :hasSymptom :A} with a True context.
> {:John :hasSymptom :B} with a False context.
>
> Then, to write some custom rules for inferring stuff based on the
> context/graph to which each triple belongs.
> But i couldnt find any reference to context-aware inferring or how to
> describe such rules.
>
> Is this possible, is it in the works, is there an alternative to model the
> explicit negation and draw conclusions based on it?
>
> 2.Mutual exclusive properties.
>
> Does anyone know if there's any OWL construct or modeling pattern for
> mutually exclusive properties?
> Something like
>
> A and B cannot be in both the relations P1 and P2. For now i use a custom
> rule like:
>
> A P1 B
> A P2 B
> P1 <:MutuallyExclusive> P2
> ----------
> A <:InconsistentRelationWith> B
>
> Is there a special construct or pattern in OWL that already does this?
>
> Thanks,
> Robi
>
------------------------------
_______________________________________________
OWLIM-discussion mailing list
OWLIM-discussion@ontotext.com
http://ontotext.com/mailman/listinfo/owlim-discussion
End of OWLIM-discussion Digest, Vol 25, Issue 4
***********************************************
------------------------------------------------------------------------------
_______________________________________________
OWLIM-discussion mailing list
OWLIM-discussion@ontotext.com
http://ontotext.com/mailman/listinfo/owlim-discussion
medical.ttl
Description: Binary data
_______________________________________________ OWLIM-discussion mailing list OWLIM-discussion@ontotext.com http://ontotext.com/mailman/listinfo/owlim-discussion
