[Pellet-users] ranking in matchmaking..

Mon May 14 19:26:08 UTC 2007

Hi, Krishna...
    It looks like you're using approximately the approach I demonstrated
with my example a while back, so I've attached an extended version that
example that demonstrates a possible solution for ranking.  The solution
makes use of the "punning" feature of OWL 1.1, which allows you to have
a class, property and individual (or any combination of those) all with
the same name, but having no effect on the semantics of each other.
Having the same name, though, allows you to tie them together at the RDF
level, so queries such as with SPARQL can integrate data across these
lines.
    As before, the example job has qualifications of 1) Person must be a
citizen of Vietnam or Thailand, 2) Computer Science degree and 3) skill
in Object Oriented programming.  Each of these qualifications is modeled
with a separate class (citOfVOrT, degCS and sklOO, respectively).  I've
chosen to make the citizenship qualification an absolute requirement,
such that those persons who are not citizens will not be considered at
all.  Therefore, the class representing this job, jobCSOOVOrT, is made
equivalent to the class representing the citizenship qualification.  The
ontology uses intersectionOf, for this, even though there is only the
one required qualification, but I left it in to show how multiple
qualifications could be included.
    For the other qualifications (CS degree and OO skill), I created a
Qualification class whose members will be individuals with the same name
as qualification classes (degCS and sklOO, in our example).  A
floating-point weight value must be specified for each desired
qualification.  This allows emphasis to be placed on certain
qualifications over others when ranking CVs.  Each job class will have a
datatype property created which will be a sub-property of
qualificationWeight.  Each desired qualification will use this property
to specify the weight for that qualification.
    This sub-property mechanism is a means of getting around the lack of
higher arity relations in OWL, as the property itself carries the
knowledge of which job class it relates to.  To make this knowledge
explicit so that the query can operate on it, a hasQual property relates
an individual with the same name as the job class to an individual with
the same name as the datatype property.  As a side note, a similar
technique could be used to, for example, specify for a CV the number of
years of experience in a particular skill.  A base datatype property,
such as hasExperience, would setup the domain and range, then a
sub-property of that for each skill would be used to specify the value
for an individual CV.  Another property, similar to hasQual in our
example, would relate the skill to an individual with the same name as
the property that represents experience in the skill.  OWL 1.1's
user-defined datatype mechanisms could then be used to define
qualification classes for different ranges of experience in a given
skill.
    Back to the example, the following query will produce results
listing a person (CV), job, qualification and ranking weight value for
those who meet the minimum requirements for the job and some additional
qualifications.

PREFIX rdf:  <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
PREFIX jobs: <http://www.owl-ontologies.com/jobs.owl#>

SELECT ?pers ?job ?qual ?weight
WHERE { ?pers rdf:type jobs:Person .
        ?job rdfs:subClassOf jobs:Job .
        ?pers rdf:type ?job .
        ?job jobs:hasQual ?prop .
        ?qual ?prop ?weight .
        ?pers rdf:type ?qual }

    Summing the weight values for each combination of CV and job will
produce a ranking.  I think this is probably the least
programming-intensive way to do something like this, since OWL (and
therefore Pellet) doesn't provide a way to talk about partial matches to
a class, nor a reasoner would decide which partial match is better than
another.  Of course, there's probably some good papers out there about
general approaches and how such a mechanism could be added directly to a
language like OWL, but that's a little beyond my scope, at the moment.
:)
    Note that this query and the attached ontology have been tested with
Pellet 1.4.  I'm fairly sure this version is required, as it provides
the support for OWL 1.1 punning.  On that note, though (hey, Evren!), I
did get the following warnings when running this query at the command
line:

Non OWL-DL features used:
Multiple Types: Resource jobs:degCS  is defined both as Class and
Individual
Multiple Types: Resource jobs:jobCSOOVOrT  is defined both as Class and
Individual
Multiple Types: Resource jobs:sklOO  is defined both as an individual
and a class
Multiple Types: Resource jobs:hasQual  is used as an individual but
defined as aclass

    I assume this is just Pellet not handling species determination in
the presence of OWL 1.1 feature use very well (a known problem in 1.4)?
Naturally, the use of punning DOES make this non-OWL-DL, so the warnings
are correct.  However, the last one struck me as odd, since hasQual is
not used as an individual, but does cause jobCSOOVOrT-q, which is
defined as a datatype property, to also be used as an individual.
    I hope this example and explanation are reasonably clear and
helpful.  Kudos to the OWL 1.1 gang for including punning in the new
features!

-Brandon :) 

________________________________

From: pellet-users-bounces at lists.owldl.com
[mailto:pellet-users-bounces at lists.owldl.com] On Behalf Of Krishna
Sent: Monday, May 14, 2007 1:01 PM
To: Pellet-users at lists.owldl.com
Subject: [Pellet-users] ranking in matchmaking..

Hi all,
After getting some idea from this group and some study, I have learnt to
match job with CV. In my ontology  I have CV as individual of class
PERSON and jobs as subclasses of JOB class. with this approach I can
only decide if certain CV matches with a job or not. But I can't rank
them. 

If I had modelled CV as subclasses, I  could have divided them into
match classes( exact, full, plugin,  intersection or disjoint). But
still I can't make rank of CVs? can I do so?

Is there someway to do it (with some coding)  or what can be  done for
getting rank in a same match group ?

Thanks in advance.
Krishna

________________________________

Yahoo! oneSearch: Finally, mobile search that gives answers
<http://us.rd.yahoo.com/evt=48252/*http://mobile.yahoo.com/mobileweb/one
search?refer=1ONXIC> , not web links. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.owldl.com/pipermail/pellet-users/attachments/20070514/aa26bbf8/attachment-0001.htm 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: jobs.owl
Type: application/octet-stream
Size: 4191 bytes
Desc: jobs.owl
Url : http://lists.owldl.com/pipermail/pellet-users/attachments/20070514/aa26bbf8/attachment-0001.obj 