Page tree

Versions Compared


  • This line was added.
  • This line was removed.
  • Formatting was changed.



updated timeline
timeline approved

proposed ARK spec cleanup 
no objections to broad cleanup of spec, even if they generate noisy diffs

ongoing ?info inflection proposal discussion

  • json vs json-ld (rdf)
  • accommodating descriptive hierarchy
    • finding aid > box > folder > letter
    • dataset > row > cell
    • article landing page > pdf file

KH: article seems to still endorse RDF a bit; there's value in aligning with DataCite
BC: going from XML to JSONLD logic isn't just a question of applying an xslt; we spent several years to come up with xml / mets data model; Gautier Poupeau (author of blob post) designed the dig repo at BnF
BC: agree that for people who have XML-based metadata, it's easy to express it as json; but a json-ld expression requires revising your whole data model and so is much harder and very complicated
GJ: could you explain why it's not just a simple mapping?
BC: because you have to express everything in triples with entity-relationships
JK: why isn't it just a set of key/value pairs?
JK: should we invite Justin L to future subgroup meeting to help us understand? Bertrand, do you want to join as well?
KH: yes, that would find that helpful
BC: yes, I would
GJ: see
BC: for persistence, we'll be defining our own vocabulary
BC: if we went with Dublin Core entity-relationship model, that's simple enough; but if they have to invent their own model, that's much harder; by using linked data, one says that all my data is expressed as triples
KH: see also
BC: nice to have a position about how we express metadata, but we have to go further and say, at least for basic metadata, what the recommended model is
JK: *model* is important – we need to understand this better as a group
KH: fortunately, vocabulary choice is independent of syntax
BC: at BnF we chose XML and snippets; we could recommend a set of optional return formats; users at many French archives will have a hard time putting out something different from what they already do; this is an argument for giving people options for returned data
KH: make it lightweight, but steer people in certain directions
KH: See the Portland Common Data model
SM: Portico would like a conceptual model for a landing page as an archival unit; it would be an enormous simplification to be able to have a landing page model for this


In follow-up email from KH:

Reflecting on the discussion, I think it may have gotten lost that when we were discussing JSON-LD a few weeks ago, I believe it was imagined as part of a series of recommendations, not as requirement. If that is still the case, then perhaps it’s not necessary to add a meeting to discuss the pros and cons of linked data further, since implementation would be looser and JSON-LD would be encouraged but not required.  I think it should be perfectly fine for Portico to produce JSON-LD and BNF to produce XML.

I’m wondering if the guidelines might go something along these lines:

  1. At minimum, ?info must resolve to a human readable landing page, and should provide a gateway to machine-readable metadata
  2. It is strongly recommended that meta tags with [something like] DC are implemented (since they are simple html, and all orgs should be able to do something with those).
  3. Secondary to this, we encourage but don’t require JSON-LD with (and/or DC) on the ?info page (in alignment with the JDDCP recommendations in the Scientific Data article). 
  4. Finally, regardless of whether JSON-LD is implemented, we encourage organizations to use whatever data format[s] is appropriate in their context as the machine-readable data version of ?info, but encourage that:
    1. Organizations include DC metadata in this where possible
    2. Organizations utilize either content negotiation or add “&format=[json|xml|etc]” property to deal with alternative formats.

This is just a rough example, but maybe something like this approach might work to give a little structure but plenty of flexibility.

Action items

  •  make sure ?info leaves options for people, but steers them towards json
  •  add Bertrand to subgroup