Consider supporting RDF-star and JSON-LD-star for temporal aspects of properties (and deprecating Role)

[mfhepp]

RDF-star is a well-defined mechanism for describing the temporal aspects of a property or sets of properties (and more). There is now work on a proper specification for RDF-star in JSON-LD syntax called JSON-LD-star.

We have several properties in schema.org that are used to attach validity statements and other meta-data to an object or a related RDF node for an n-ary relationship, like https://schema.org/validFrom.

I propose to consider using RDF-star as the future mechanism of choice for validity statements and other meta-data.

For a detailed argument, see #3098 (comment)

Required tasks:

• JSON-LD-star specification must be finalized and adopted
• Major consumers of schema.org markup must be able to process JSON-LD-star
• Decide which schema.org properties should be allowed for annotation objects
• Update examples and guidance

[gkellogg]

The JSON-LD-star document referenced is out of date now, given that the design of RDF-star has evolved. It will need to consider some of the new concepts (Triple Terms vs Quoted Triples and indirect reifiers (see w3c/rdf-concepts#80)), but it should make it into a future JSON-LD 1.2, along with other things that have been evolving, maybe a year after RDF 1.2 Concepts goes to REC.

[mfhepp]

Thanks, @gkellogg !

But do you agree with my general proposal to be very reluctant to add properties for temporal and other meta-data and plan for the use of RDF-star in the future?

At least we should keep this future use-case in mind when adding or updating existing properties like validFrom etc. - they may work very well when attached to a triple.

[gkellogg]

There is some discussions in the RDF-star UCR issues about temporal relationships, but more warnings than anything else. Of course, we welcome more specific use cases that touch on this need.

I think that RDF-star provides a useful structure for adding such assertions, and is the right way to think about this. There's also open discussion on if a reifier identify more than one triple term, or if there is a cardinality constraint. This could allow for inferred triples to be associated with the same identifier, but could potentially be used beyond that, and start to blur the distinction between named graphs and reified triples. Opinions on this from the community are encouraged. Ultimately, the purpose of adding such features to the RDF core data model need to support real-world use cases.

Regarding what schema.org should do, I would tend to support waiting for RDF-star features before going beyond what is already afforded using Role.