Use case: Dataset size characteristics

[VladimirAlexiev]

Submitting a new USE CASE:

---

Dataset subsets and size characteristics

Status:

Identifier: ID51 (proposed)

Creator: Vladimir Alexiev, Ontotext

Deliverable(s): DCAT1.1

Tags

semantics statistics size

Stakeholders

Data consumers often need to know how many of what sort of entities are included in a dataset.
In an aggregation scenario, different subsets (parts of a dataset) need to be expressed, eg because they come from different data providers.

Eg in the euBusinessGraph project we have a need to describe Company datasets by different providers,
what properties are included in each (eg ebg:isStartup, org:orgActivity),
and some partition info eg "the dataset covers jurisdiction Italy" or "the dataset has 1000 Italian startups"
(i.e. rov:RegisteredOrganization with ebg:isStartup=true and jurisdiction Italy)

Problem statement

DCAT 1.0 has only a property dcat:byteSize, which is pretty useless to describe any aspect of dataset content or value.

And it has no means of expressing subsets.

Existing approaches

VOID statistics includes these void: counting props: triples, entities, classes, properties, distinctSubjects, distinctObjects, documents.

Very importantly, these can be used on subsets such as classPartition and propertyPartition, which provides very powerful means to describe exactly what kinds of entities are present, and how many are in the dataset.
Thus I believe that subsets are instrumental in expressing the fine-grained content of a dataset.

Links

Schema issue https://github.com/schemaorg/schemaorg/issues/1855

Requirements

Ability to express the fine-grained content of a dataset:

• Ability to express subsets of a dataset.
• Describe subsets by kind of entity (e.g. Companies vs Events) and/or entity characteristics (e.g. Italian companies, Startups, Startups in Italy)
• The kinds and characteristics should be expressed by URLs
• Express the count of entities in a dataset or subset
• Optionally, express other dataset size characteristics. E.g. in RDF context, that's number of triples and nodes

Notes:

• It's pretty clear how to do this for RDF datasets (see VOID). The real challenge is how to do it for other datasets.
• I think it's mandatory to express subset characterization with URLs and not text.
• I think there's also need to express assertions to be used for characterization, eg ebg:isStartup=true

Related use cases

ID33, ID7, RDSAT, RSS.

This one could be merged into ID33 to provide further details.

[dr-shorthair]

This looks like a clear gap in DCAT capabilities.
Primarily applies to Dataset, but also to Catalog

Could UCR team extract Requirements from this?

[andrea-perego]

@dr-shorthair , @VladimirAlexiev ,

I think this use case relates very much to the following reqs:

• Related vocabularies comparision [RVC] Related vocabularies comparision [RVC] #87
• Related vocabularies mapping [RVM] Related vocabularies mapping [RVM] #88

and also to the following UC:

• 5.44 Identification of versioned datasets and subsets

[VladimirAlexiev]

So both this one (ID51) and 5.44 (ID44) talk of subsets but also about other topics.
I feel that subsets are such a crucial topic, they should be split into their own requirement.

[jpullmann]

The proposal seems to merge two concerns: 1) expressing the structure of data sets, like modeled in DATS (hasPart) and 2) indicating the size property of the composite or the individual parts. My suggestion is to rephrase this UC to cover only 2) and create and link here a new UC on behalf of 1)

[VladimirAlexiev]

hi @jpullmann! I've reread this proposal and I agree with you. As I wrote "subsets are such a crucial topic, they should be split into their own requirement". Requirement 5.44 (ID44) also talks of subsets, but for another reason.

Someone needs to take subset characterization parts from this one (ID51), 5.44 (ID44), and your suggestion (DATS hasPart) and consolidate them. I could try it but I'm not sure I can capture other people's suggestions adequately. Are there designated editors in this WG?

[dr-shorthair]

Editors names on each draft https://www.w3.org/2017/dxwg/wiki/Main_Page#Deliverables
For UCR document it is Ixchel Faniel, Jaroslav Pullmann, Rob Atkinson.

[dr-shorthair]

Relationship of a dataset to subsets is part of #81
This issue should focus on better description of dataset size, perhaps including extent on different dimensions.

I've renamed it to reflect that focus.

[agbeltran]

As per the last comment, assuming that the related datasets (sub-datasets) are considered in the requirement #81, should this UC be added to the UCR document focusing on the dataset size characteristics where the main requirements would be around (copied from above):

• Express the count of entities in a dataset or subset
• Optionally, express other dataset size characteristics. E.g. in RDF context, that's number of triples and nodes

Ping to @fanieli @jpullmann @rob-metalinkage

[davebrowning]

Also related to existing issue/requirement #84 and some of the discussion of #313.

[dr-shorthair]

@VladimirAlexiev are you interested in reviving this?

[andrea-perego]

Unless there are any objections, I propose we close this issue.

[andrea-perego]

Noting no objections, I'm closing this issue.