dcterms:subject (free-text) about gcis

[bduggan]

1. Yes, I believe t is possible to construct such a SPARQL query.

[justgo129]

I'd like to reopen this but don't have the permissions. I'l provide the rationale upon the reopening (it would make it easier to follow on my part).

[bduggan]

What is the rationale?

[justgo129]

@zednis is it a problem for dcterms:subject to contain multiple objects?
Basically, it's an extension of the discussion in #150 where a similar discussion occurred vis a vis output file names (i.e. splitting them up in the turtle or not).

Once I gain some clarification on it, I am all right with closing.

[zednis]

Two things:

1. We should avoid collapsing multiple values into single literals. The property is not intended to reference an embedded collection but a single value. The correct way to do this in RDF is to make multiple RDF statements reusing the property with different values.

2. dcterms:subject is intended to be used as an object property and not with literal values.

from http://dublincore.org/documents/dcmi-terms/#terms-subject

This term is intended to be used with non-literal values as defined in the DCMI Abstract Model (http://dublincore.org/documents/abstract-model/). As of December 2007, the DCMI Usage Board is seeking a way to express this intention with a formal range declaration.

Most terms in dcterms are intended to be used with non-literal values. This is what separates dublin core terms properties from regular dublin core properties - which are intended to be used with literal values.

Guidance on usage at available at http://wiki.dublincore.org/index.php/User_Guide/Publishing_Metadata#dcterms:subject

The correct representation in RDF for dcterms:subject would be like this:

<http://data.globalchange.gov/image/ff6a7a8e-d886-4b30-acd7-a3538a787baf>
  dcterms:subject <http://data.globalchange.gov/term/Precipitation> ,
                            <http://data.globalchange.gov/term/observed>,
                            <http://data.globalchange.gov/term/global> .

This is why dcterms:subject is usually used to reference an instance of a skos:Concept.

The correct representation in RDF for dc:subject would be like this:

<http://data.globalchange.gov/image/ff6a7a8e-d886-4b30-acd7-a3538a787baf>
  dc:subject "Precipitation"^^xsd:string, 
                   "observed"^^xsd:string, 
                   "global"^^xsd:string .

[zednis]

@bduggan this issues should be reopened and we should modify your usage of dcterms:subject because it is incorrect.

[bduggan]

I would propose removing dcterms:subject from the rdf. This falls into the larger category of controlled vocabularies which is work in progress. Again, this data was provided as a free form text field, and we do not want to make any assertions about splitting the contents and making URIs for various phrases there.

[zednis]

I think this information is too important to just drop. What if we have a student work on a post-process to split free-text provided contents into separate text keywords. We can then use either dc:subject or dcterms:subject as mentioned above.

[bduggan]

The lists in the attributes are not constrained, and there are no definitions. An example: "Precipitation, projections, seasonal, CMIP5, RCP2.6" -- collections of keywords and a model and a scenario. I would argue that treating these as dcterms:subject's is not a good idea -- better would be for the image to be associated with /model/cmip5 and /scenario/rcp2.6. Once we start properly managing controlled vocabularies we will want definitions for each of the terms as well as how they relate to other controlled vocabularies.

I think these lists are important to keep in mind and should help inform our discussions about representing (and curating) controlled vocabularies.

[zednis]

I don't want to just throw this data out.

I think it makes sense to use dc:subject in the meantime before we have controlled vocabularies fully in use - and it will allow us to support keywords that don't map to a term in a controlled vocabulary.

We should do some data science to post-process the free-text data we have into delimited keywords, but thats why we have students ;-)

[bduggan]

I'm not so worried about the effort of splitting these up, I'm more worried about making a category mistake with this data. Besides models and scenarios, there are also regions and even platforms/instruments: things for which we already have URIs.

Anyway, eventually, yes, we do want URIs for term, perhaps under "/term" (in your example)., To support this, we need to decide what is returned from the /term endpoint. i.e. what are the attributes and relationships of a term? (not in the lexicon sense but in the controlled vocabulary sense.)

Also, I wasn't suggesting throwing it out, just leaving it out of the RDF for now; it'll still be in the database and we can add it once we iron out our representation of controlled vocabularies.

[aulenbac]

@bduggan is right. Think engineered. Think lean and automated. And think of an operational GCIS ontology that is automatically generated on a weekly schedule using that lean automated approach. Manual work is not an option; we do not have students for new work like this nor should we assume we shall in the future.

Actually, I am okay with throwing this "data" out. Based on two years of GCIS work, Attributes is a spotty, usually blank, catch-all that few bother to fill out. When they do, the values are amazingly varied and not terribly useful. Perhaps we should remove the Attributes field completely.

[justgo129]

I'd be all right with removing the "attributes" field completely given the lack of criteria for the process of assigning them. Let me confirm first that this is so and will update.

[zednis]

ok, I have been persuaded to the leave this content out of the RDF for now. I support Brian's suggesting of revisiting this and eventually bringing it back with controlled vocabularies.

[justgo129]

Works for me.

On Wed, Jun 24, 2015 at 2:26 PM, Stephan Zednik [email protected]
wrote:

ok, I have been persuaded to the leave this content out of the RDF for
now. I support Brian's suggesting of revisiting this and eventually
bringing it back with controlled vocabularies.

—
Reply to this email directly or view it on GitHub
#175 (comment).

---

Justin Goldstein, Ph.D.
Advance Science Climate Data and Observing Systems Coordinator
US Global Change Research Program
1717 Pennsylvania Ave NW, Suite #250
Washington, DC 20006

O: [phone]
M: [phone]

e-mail: jgoldstein AT usgcrp Dot gov
http://www.globalchange.gov