Re: purpose/goals for observations ontologies

Hello John, comments inline,

On Tue, 2009-08-04 at 13:52 -0600, John Graybeal wrote:
> On Aug 4, 2009, at 10:20 AM, Kevin R. Page wrote:
> > We should recognise that both user-oriented (data) and process- 
> > oriented
> > (sensor) use cases exist (as reflected in current OGC standards).
> I am having trouble with this framing; maybe just an ambiguity, or  
> maybe more.

So I know there are those on this list who are more familiar, and can no
doubt elaborate more eloquently, on the distinctions made in current OGC
standards - please do (and correct me :)  )

I guess my bracketed 'data' and 'sensor' above show where I see the
differences (and I don't want to overdo them as differences).

I'll start with an (over-simplistic) description of where I'm coming

1) sometimes, we might start with a sensor network, with it's elements
described according to the device ontology. We might use the ontology to
manage the sensor network. We might use the descriptions of sensor
properties and data capabilities to pick out particular sensors, and
from there get to the data that sensor has produced. Absolutely, this is
the device ontology.

2) at other times we might start with a large amount of data produced by
a sensor network, and from that we want to create useful information.
It's more than just data; we care about concepts like observations,
measurements, context, so that we can process the data effectively.
Descriptions about the actual sensors is metadata to this data; that's
not so say it isn't important, it very much is (e.g. as provenance, or
to infer the classification of the data from the sensor capabilities),
but we're starting from the data.

I don't think there's any horrific difference or schism here. There's
obviously overlap - it's the same data. Sometimes you come at different
parts of it from different directions.

And it's much easier to bring these two viewpoints together in the RDF
world than the XML Schema world.

So a device ontology might have some O&M concepts included; an O&M
ontology might have some device concepts included; it might be one big
ontology (don't have to use all of it, after all).

As long as whatever ontology (or ontologies) we end up with enables us
to just have devices, or just have observations, and get from one to the
other as and when we can (or want to) link that data.

>From another perspective: semantic web technologies can be applied to
improve sensor networks; but I think it's equally, if not more,
important that sensor networks and the data they output become part of
the semantic web of data. These aren't orthogonal tasks.

> I agree that use cases about the (actual output) data *produced by*  
> sensors exist.

and it matters that this data was produced by sensors; these use cases
need to capture and encode this.

> Use cases about the data *describing* actual sensors  
> (name, size, color, and all that) also exist. The latter is what I  
> thought a device ontology should encompass.

Yes. And perhaps 'device ontology' is a clearer description of that
ontology if it doesn't include O&M concepts.

> So, which of these did you mean by 'user-oriented (data)'?  (I suggest  
> that 'user-oriented' is entirely a function of the user, and some  
> users care only about the devices, not their data; so maybe this isn't  
> an optimal term.)

Indeed, I am not fond of the term.

So I think 'user-oriented (data)' as originally cited is the former -
but the data describing sensors is still there as (vital) metadata.

(Illustrative use of the term 'metadata' - I'm not sure I believe in
metadata enough to classify what is and isn't data ;)  )

> Will the introduction of the 'process oriented' way of looking at the  
> device -- the framing introduced by SensorML, which I have heard  
> summarized as "the sensor is a process", right? -- tell me more, less,  
> or the same information as a 'simple descriptive model'?

About the device? The same. I think the 'process' concept encapsulates
the manner by which the observation was gathered. When this is a sensor,
the information about the 'process' instance is (or could be) the simple
descriptive model / the device ontology.

> Put another way, is there necessarily any difference between the two?

I'd rather there not be. I think we can do both.

As Krzysztof's recently arrived email says, a good starting place is
probably to extend the observation concept in the sensor ontology.

> To tie this back to the larger question I started with, It just seems  
> to me that where some element comes from a process, the ontology will  
> naturally describe that ("sensor producesDataRecord recordType1").

And when I come across an instance of 'recordType1' I want to know that
it was produced by an instance of 'sensor'.



Kevin R. Page  
Intelligence, Agents, Multimedia      University of Southampton, UK

Received on Tuesday, 4 August 2009 21:54:57 UTC