search

w3c / sdw-sosa-ssn

Repository of the Spatial Data on the Web Working Group for the SOSA/SSN vocabulary
7 stars 5 forks source link

Sample modeling in sosa/ssn #19

Open croussey opened 5 years ago

croussey commented 5 years ago

we try to reuse the sosa ontology to design a contextual system about irrigation. We notice in the sosa documentation that the class Sample is defined as a subClass of FeatureOfInterest... http://w3c.github.io/sdw/ssn/#SOSASample not that in sosa ttl file Sample class is not defined as a subClass of FeatureOfInterest Class.

I found the subClassOf modeling very strange because the link between a sample and a featureOfinterest is a kind of partOf link... The sample maybe a part of the feature of interest where the observation is made.. sample express that the feature of interest can not be taken as a whole to make an observation... It is really true about weather observation or soil observation When the featureOfInterest is small I could understand that a sample is one of the instance of the featureOfInterest but it is not the case all the time..

The property to describe that a sample is part of a feature is isSampleOf (with the inverse hasSample).

The main point is if a sample may be linked to a featureOfInterest by a partOf relationship (isSampleOf) thus you should remove the subClassOf relationship between a FeatureOfInterest class and Sample class. Is it possible to remove this information from the sosa documentation?

Best Catherine Roussey catherine.roussey@irstea.fr

dr-shorthair commented 5 years ago

Hi @croussey

  1. sosa:Sample is a subclass of sosa:FeatureOfInterest because every sample is expected to (eventually) be the feature-of-interest of a sosa:Observation, else it is of no interest

  2. There are no rdfs:subClassOf axioms in sosa.ttl - the axiomatization in this graph was deliberately limited to weak schema.org style (but see [1]). The subclass axioms are in ssn.ttl.

  3. Subsumption and mereology are different kinds of relationship. The subsumption (subclass) relationship is general - it applies to all members of the class, i.e. 'every sample is also a feature-of-interest'. On the other hand, mereological relationships (part/whole) are between individuals, e.g. 'this individual sample is part-of that individual feature-of-interest'. [2]

  4. The point of samples is that they provide a tractable thing for observations, but with the express intention that the result can be used (somehow) to characterize the thing that was sampled. A Sample is the proximate feature-of-interest of an observation. The thing that was sampled is the ultimate feature-of-interest of the investigation.

[1] We made an error in using owl:inverseOf. This should have been schema:inverseOf but we realised that too late after the ontology had already been published.

[2] We can axiomatize the definition of 'partOf' to say things like 'a partOf relationship always relates a sample to a feature-of-interest'. But note that this only says 'if this relationship is present, then the individual on one end is a member of the class of samples, and on the other is a member of the class of features-of-interest' (which might also be a sample)

croussey commented 5 years ago

So feature of interest is just a role... the subject of an observation. My question is why do we need a class to express this role? because anything can be observed ... an observation maybe the subject of another observations...There is not constant characteristic that differentiate the subject of an observation to anything else. I was expecting that feature of interest express a kind of topic to classify observations: medical observation, earth observation, weather observation. Each of this domain has defined a list of reference topics that they are interested in... this topic are what you call the ultimate feature of interest... for each ultimate feature of interest each domain will defined a list of properties able to be observed.

kjano commented 5 years ago

Hi Catherine,

I am not sure whether I understand your point. The fact that FeatureOfInterest can be modeled as (thematic) role does not suggest not to model it in an ontology (as a class). You may, just as a first example, be interested in every water body for which the water level has been measured since the last rain. You may also be interested in creating subclasses of FOI. In yet another context, you may want to tell apart the feature of interest from the ultimate feature of interest, and so on.

Best, Jano

dr-shorthair commented 5 years ago

Yes, feature-of-interest is primarily a role.

We also wanted a class for the set of things on which observations had been, or might be, made. You are quite correct that this could be more or less anything, so the sosa:FeatureOfInterest class is very very general - similar to GF_Feature or Feature from ISO 19109. We could possibly have just used owl:Thing or owl:NamedIndividual, but we went with sosa:FeatureOfInterest.

As @kjano implies there is no harm in giving this class a name, and it allows us to add additional axioms which assist in clarifying some of the intentions elsewhere in the ontology (e.g. around sampling).

nicholascar commented 5 years ago

I have found, though work with scientific data collection organisations, that it's very useful to have the sosa:FeatureOfInterest class specifically named. It's good to specifically call out the target of observations to ensure that the organisation can articulat what it's actually collecting observations/results/samples for. Also, use of the ssn-ext:hasUltimateFeatureOfInterest property indicating a sosa:FeatureOfInterest has been useful as sometimes the observations/samples etc. are themselves of another sampe of the ultimate feature of interest and yet we want to link all samples etc. to the things we actually want to know about - the real features of interest.

lieberjosh commented 5 years ago

I wouldn’t want to undervalue the importance of identifying sosa:FeatureOfInterest as being of the physical world, yet with ObservedProperties that result from an observation process. That act of interpretation and commitment is very commonly overlooked, especially in IoT / WoT circles, sometimes with tragic results, c.f. Lawrence, MA gas pipe explosions.

—Josh

ronojinx commented 3 years ago

The explanation that FeatureOfInterest is a role now makes it very understandable, I am importing SOSA/SSN and was wondering about the need of sample. The thread has explained it quite well @dr-shorthair , thanks!!!

dr-shorthair commented 10 months ago

Perhaps some of this discussion needs to be folded into the spec?

dr-shorthair commented 1 month ago

Check to see if the concerns here are reflected in the SSN Ontology documentation, in particular in

and maybe also in