Is it possible to specialise dcat:keywords or dcat:theme and use a custom controled vocabulary?

[aisaac]

Context:

Following advice I'm posting a question here... This is about using DCAT-AP in the work of a Europeana Network working group on Datasheets for digital cultural heritage. We have defined a template to describe datasets from our domain (https://zenodo.org/records/15828222) and are working on aligning it with DCAT-AP.

Question:

One requirement we have is to capture the "Category of AI usage" and "Type of CH Application" that a dataset could be used for. We thought of creating sub-properties of dcat:keywords but this property has a range of literal and we'd like to use it with URIs from a controlled vocabulary.

We also thought of specialising dcat:theme. My colleague @nfreire has found this: https://semiceu.github.io/DCAT-AP-reuse-guidelines/#situation-controlled-vocabulary-mismatch , which hints that for property assigned with a vocabulary there could still be an option to using additional statement with a different vocabulary. But it seems that in DCAT-AP all statements have to use the NAL:data-theme vocabulary ("The values to be used for this property are the URIs of the concepts in the vocabulary." at https://semiceu.github.io/DCAT-AP/releases/3.0.0/#controlled-vocabularies-to-be-used). So we would have to create our sub-properties as entirely new properties not specialising existing DCAT ones.

Can you confirm this is the only (foreseeable) route for such requirements? Or have I misread the documentation (and thus it should be clarified by a MUST maybe)?

[EmidioStani]

Hello @aisaac ,

I jump in this issue as it is related to AI.
You might know that we are working MLDCAT-AP and I see in your paper there is a need to describe "Papers and Other References", "AI Category", "Citation Information", "Discussion of Biases", so we could have a separate discussion on this topic. What do you think ?

[H-a-g-L]

@aisaac, Did you consider using vann:usageNote? The GeoDCAT community had a similar discussion looking for an equivalent property for mapping of gmd:purpose. In the cited issue, the geodcat:purpose property was created, because the expected range is a Literal (while vann:usageNote has a range of rdfs:Resource).

[aisaac]

Thank you for the feedback! @EmidioStani yes we could have a separate discussion. I'm sending you an email.

@H-a-g-L we didn't consider vann:usageNote, because we wanted to first try to find something in DCAT-AP and vann:usageNote is not in DCAT-AP, right?
Now if it's impossible of dcat:keyword we'll look at alternatives

[H-a-g-L]

we didn't consider vann:usageNote, because we wanted to first try to find something in DCAT-AP and vann:usageNote is not in DCAT-AP, right?

No, it is not part of DCAT-AP but is included in the DCAT-AP-JRC extension. However, vann may not be the best fit for the specific use-case so I would be interested to see if ML-DCAT-AP is able to support the need to document usage and restrictions of resources in ML context.

[EmidioStani]

Hello @H-a-g-L ,

so far "how" a dataset is used in ML context is expressed by the relations between MachineLearningModel and Dataset, such as "trained on", "tested on", "validated on" and "runned on".

Being a DCAT-AP based profile, a Dataset also has dcat:keyword and dcat:theme for classifying the dataset but no code list is recommended.

In the AI Code Of Practice there is a transparency requirement for the data, described as:

Scope and main characteristics:
A general description of the scope and main characteristics of the training, testing and validation data, such as domain (e.g. healthcare, science, law,...), geography (e.g. global, restricted to a certain region,...), language, modality coverage, where applicable. [Recommended 200 words]

Which, so far, is translated in Dataset description, theme, geographicalCoverage and language aside the relations mentioned at the beginning .

[LudgerRinsche-initAG]

You can use theme. Starting with 3.0.1, dcat:theme has only to be used "with AT LEAST 1 value from the controlled vocabularies": https://semiceu.github.io/DCAT-AP/releases/3.0.1/#properties-with-at-least-1-value-from-the-controlled-vocabularies

[JustasTamulis]

You can use theme. Starting with 3.0.1, dcat:theme has only to be used "with AT LEAST 1 value from the controlled vocabularies": https://semiceu.github.io/DCAT-AP/releases/3.0.1/#properties-with-at-least-1-value-from-the-controlled-vocabularies

Hello, I am trying to extend the dcat:theme as well and I found this statement to be incorrect. The AT LEAST 1 is applied per node, not per dataset, therefore each dcat:theme field must be from controlled vocabularies.

This can be checked by running provided shacl shapes: https://semiceu.github.io/DCAT-AP/releases/3.0.1/html/shacl/mdr-vocabularies.shape.ttl or online in https://www.itb.ec.europa.eu/shacl/dcat-ap/upload

Minimal example of working (passing the validation) metadata in turtle:

@prefix dct: <http://purl.org/dc/terms/> .
@prefix example-ds: <https://data.gov.gr/id/dataset/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
example-ds:BeePopulation a dcat:Dataset;
  dct:title "Bee population"@en;
  dcat:theme <http://publications.europa.eu/resource/authority/data-theme/AGRI> .

Producing a warning:

@prefix dct: <http://purl.org/dc/terms/> .
@prefix example-ds: <https://data.gov.gr/id/dataset/> .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .

example-ds:BeePopulation a dcat:Dataset;
  dct:title "Bee population"@en;
  dcat:theme <http://publications.europa.eu/resource/authority/data-theme/AGRI>,
    <http://inspire.ec.europa.eu/theme/hh> .