W3C home > Mailing lists > Public > public-prov-wg@w3.org > September 2011

Re: Roles (was: Testing the ontology for expressing workflow provenance)

From: Graham Klyne <GK@ninebynine.org>
Date: Thu, 15 Sep 2011 12:14:05 +0100
Message-ID: <4E71DDFD.8030104@ninebynine.org>
To: Simon Miles <simon.miles@kcl.ac.uk>
CC: Provenance Working Group WG <public-prov-wg@w3.org>
Short answer: yes.

...

Slightly longer answer:
Depending on the toolkit being used, specializing properties can sometimes make 
it harder to extract generic information from a speclialized rendering.

E.g. if we have

   somedomain:fooInputData rdfs:subPropertyOf prov:used

and then use just somedomain:fooInputData in some provenance data, it may be 
harder for a non domain-aware application to be aware of the prov:used relation 
implied by somedomain:fooInputData, in that the application may have to employ 
something like an RDFS inference engine to expose this.

#g
--

On 15/09/2011 11:02, Simon Miles wrote:
> Hello all,
>
> Just to clarify, I understand that specialising used/generated to
> assert roles makes it hard for roles to have structured information,
> and that if the implied prov:used/prov:generated relation is not
> (also) included in a serialisation, non-reasoners will not be able to
> traverse the provenance graph.
>
> I was not advocating that this *should* be done, but that I expect
> people *will* do to this, especially when they are using their own
> ontology for describing domain-specific information. Specialisation of
> properties is surely the normal way to provide more specific
> information about how things are related, i.e. their roles with regard
> to each other. I agree it may not be so normal for a generic workflow
> engine, such as Taverna, where their is no pre-defined domain.
>
> The consequence of this may be just to recognise the need for guidance
> where what we are proposing does not follow the normal way of doing
> things.
>
> Thanks,
> Simon
>
> On 10 September 2011 09:45, Graham Klyne<GK@ninebynine.org>  wrote:
>> I haven't been following this as closely as I should, but I think the
>> alternative to specializing "used" may be similar to the CRM event-mediated
>> approach whereby provenance information can be incorporated with data about
>> things - extra metadata can easily be attached to an "observation" or
>> "annotation" (or similar) event.
>>
>> I think it's a good approach.
>>
>> #g
>> --
>>
>> On 09/09/2011 19:51, Daniel Garijo wrote:
>>> Hi Stian.
>>> In first place, thanks for your example. It is very helpful to get things to
>>> start moving.
>>> In second place, you are right: the ontology has not been updated yet with
>>> Satya's proposal
>>> for modeling roles. I think it is better than specializing the "used"
>>> property, since it allows
>>> adding additional information withouth transforming the "Used" property in a
>>> class (which is the
>>> way to model n-ary relationchips). If we just specialize the "used"
>>> property, then we won't be able to
>>> link the time of usage, the location of usage, or anything additional
>>> metadata.
>>>
>>> However, we are still discussing this approach, because it is true that when
>>> you don't know
>>> the role of the used entity, everything might get a bit confusing.
>>>
>>> You are welcome to join us on monday's telecons :)
>>>
>>> Best,
>>> Daniel
>>>
>>> 2011/9/9 Stian Soiland-Reyes<soiland-reyes@cs.manchester.ac.uk>
>>>
>>>> On Fri, Sep 9, 2011 at 12:35, Stian Soiland-Reyes
>>>> <soiland-reyes@cs.manchester.ac.uk>    wrote:
>>>>
>>>>
>>>>>    <
>>>> http://ns.taverna.org.uk/2011/run/2613aab1-dfe9-4a17-a4be-7589f5d388d6/>
>>>>>      a  prov:ProcessExecution;
>>>>>           prov:used [
>>>>>               rdf:type
>>>>> <
>>>> http://ns.taverna.org.uk/2010/workflow/ea4168eb-67ea-440f-ab73-818da5efc998/processor/String_constant/out/value
>>>>>
>>>>>               prov:assumedBy
>>>>> <
>>>> http://ns.taverna.org.uk/2011/data/2613aab1-dfe9-4a17-a4be-7589f5d388d6/ref/153277f1-5e4f-43fc-968d-ab3a8b038676
>>>>>
>>>>> ;
>>>>
>>>> Note that I messed up the direction here - if something was 'used'
>>>> then the role should of course be an *input* port.  Just imagine
>>>> s/Output/Input/g for the whole thing as it is not possible to edit an
>>>> email once it's sent. :-)
>>>>
>>>>
>>>> (I wanted to do the discussion on 'used' rather than 'generated' - as
>>>> use can naturally occur in several roles in several process execution
>>>> - and indeed in several roles for the same execution)
>>>>
>>>> --
>>>> Stian Soiland-Reyes, myGrid team
>>>> School of Computer Science
>>>> The University of Manchester
>>>>
>>>>
>>>
>>
>>
>
>
>
Received on Thursday, 15 September 2011 11:57:43 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 26 April 2012 13:06:41 GMT