dcat:Distributions Usage note should not encourage the usage of Distributions as Dataseries

[sabinem]

Class: dcat:Distribution Conformance Problem:

• In DCAT a distribution (https://www.w3.org/TR/vocab-dcat-3/#Class:Distribution) is defined as "A specific representation of a dataset. A dataset might be available in multiple serializations that may differ in various ways, including natural language, media-type or format, schematic organization, temporal and spatial resolution, level of detail or profiles (which might specify any or all of the above). " and in the Note below it is mentioned: "As a counter-example, budget data for different years would usually be modelled as different datasets, each with their own distributions, since all distributions of one dataset should broadly contain the same data. ", see [DCAT Class:Distribution]
• In contrast to this the current DCAT-AP CH explicitly encourages the usage of distributions to model a dataseries, see here: https://www.dcat-ap.ch/releases/2.0/dcat-ap-ch.html#Class:Distribution
• But it is also stated (https://www.w3.org/TR/vocab-dcat-3/#Class:Distribution) "Nevertheless, the question of whether different representations can be understood to be distributions of the same dataset, or distributions of different datasets, is application specific. Judgement about how to describe them is the responsibility of the provider, taking into account their understanding of the expectations of users, and practices in the relevant community. "

Details:

• In a discussion with DCAT-AP about how the Swiss property dct:coverage on dcat:Distribution can be expressed with DCAT-Vocabulary, see here: https://github.com/SEMICeu/DCAT-AP/issues/197 we were told that modeling dataseries with distributions of a dataset is considered an antipattern and the reference above was mentioned in that regard.
• In DCAT Version 3 dcat:DataSeries (https://www.w3.org/TR/vocab-dcat-3/#Class:Dataset_Series) will be added, that might be the best way to model dataseries
• Properties for distributions on the other hand include also in DCAT Version 3 only resolution but not any sort of content coverage (https://www.w3.org/TR/vocab-dcat-3/#Property:distribution_spatial_resolution, https://www.w3.org/TR/vocab-dcat-3/#Property:distribution_temporal_resolution)
• Distributions are expressed with an RDF graph, that is unordered. Therefore it is not the best fit to model a Dataseries, that usually does have an order,
• This issue is related to https://github.com/opendata-swiss/dcat_ap_ch/issues/79

Proposal:

• Follow DCAT Version 3 and DCAT-AP
• Revise the usage note on dcat:Disttribution to not encourage the modeling of a Dataseries with Distributions
• As soon as the dcat:DataSeries is available in DCAT-AP it should be added it to DCAT-AP CH
• For opendata.swiss there should be a gradual phase out of the usage of Distributions to model dataseries, that is supported by documentation on how else this can be done

[Juan-Juan-1]

So I don't know about that discussion (is it really a discussion? Wasnt it just a proposal...?) but checking both v2 and v3 of DCAT I'm not sure I understand the same thing.

First of all I would use the official definition - the statement "should all have the same content and only vary in language format or resolution" doesn't come from there (right?) and I find it to simplistic and a bit confusing...

I find this sentence super super important: "Nevertheless, the question of whether different representations can be understood to be distributions of the same dataset, or distributions of different datasets, is application specific. Judgement about how to describe them is the responsibility of the provider, taking into account their understanding of the expectations of users, and practices in the relevant community." (https://www.w3.org/TR/vocab-dcat-3/#Class:Distribution)

In general: is it really an issue which should be fixed..? Or should we just explain the concept better, but leave it to data publisher to decide?

[AFoletti]

I think we are on a difficult topic here. I for instance would personally try to avoid dataset proliferation (and thus NOT model different budget years as different datasets) because this makes in my opinion the data portal messier. But that's just a personal way of doing things.

My proposal is to explain, make examples and show best practices, but leave up to the data publisher how he manages the data.

[Juan-Juan-1]

See also https://github.com/SEMICeu/DCAT-AP/issues/155

[sabinem]

@AFoletti Please see the discussion on DCAT about this topic: w3c/dxwg#1429. Your concerns seem to be shared by a substantial part of the DCAT and DCAT-AP community.