The W3C JSON-LD Community Group

Go Back

W3C Logo

JSON-LD and Spacial Data on the Web Joint Meeting

Minutes for 2024-09-23

Present
Pierre-Antoine Champin, Gregg Kellogg, Anatoly Scherbakov, Benjamin Young, Alexandre Bertails, sebastian kaebisch, Rob Warren
Regrets
Ted Thibodeau Jr.
Chair(s)
Benjamin Young
Scribe(s)
Gregg Kellogg, Benjamin Young, Pierre-Antoine Champin
Agenda
undefined
Pierre-Antoine Champin:
Gregg Kellogg is scribing.
Rob Warren: General discussion about requirements of sensors.
Benjamin Young: Curious if issues are more ontological.
Rob Warren: Some are, but often we create standards but don't say how to use them (SOSA)
Sebastian Kaebisch: I was involved in the EXI working group (XML Exchange). They did create a primmer on how to use the model.
Sebastian Kaebisch: -> Https://www.w3.org/TR/exi/
Benjamin Young: Sometimes, standards are created, but the contributing companies otherwise don't cooperate.
Benjamin Young: We have a breakout on Wednesday to go over the JSON-LD recharter.
... Thursday, we have a focused WG meeting.
... Also with Web of Things.
... Tuesday joint meeting with RDF-star
Rob Warren: I've argued that we should ignore JSON-LD and go to content negotiation.
... The context is difficult to manage.
... I have people that try to put too much into the context.
... SPARQL Load is a problem as the context can get ignored.
Benjamin Young is scribing.
Gregg Kellogg: Yeah, dereferencing in some fashion has to happen
... Other formats is easy to get triples from, but JSON-LD requires extra steps.
... That external dependency creates problems.
Anatoly Scherbakov: Interesting discussion; this is not the only case where you need to find somethning elsewhere.
... For example owl:imports.
... I think this is an inherent part of the technology, as it is "Linked Data'.
... I agree that there are implementation problems.
Rob Warren: The jump your making is that you're going from RDF to OWL.
... I can get triples without going to OWL.
... But, with JSON-LD, I can't do anything without the context.
Benjamin Young: The concerns about JSON-LD contexts are well known.
... If you're consuming someone's JSON-LD, you don't have that option.
... Freezing contexts is a good idea, as you are not exposed to unexpected data.
... If you're building an integration, having the knowledge of what at context says is no different than referencing a specification.
Pierre-Antoine Champin is scribing.
Gregg Kellogg: Note that Google is not processing the schema.org context when they parse JSON-LD, they.
... If you use conneg to specify the format you are expecting, getting an unknown context might be seen as an error condition
Rob Warren: I deal with knowledge graphs, where you don't know the data you're going to see in advance.
... The more work you expect a person to do to read the data, the more it's going to be hard to people to bother with.
Sebastian Kaebisch: Not really a JSON-LD issue, but that of the author.
... It may be that you really can't trust the content in any case.
Benjamin Young: "JSON-LD is not JSON, because JSON can't be interpreted without prior knowledge, and JSON-LD doesn't allow that".
... JSON is easier, if you make rash assumptions and don't data model.
Pierre-Antoine Champin is scribing.
Rob Warren: I'd prefer that in the documentation we clarify that JSON-LD doesn't necessarily let you do what plain JSON can do.
Benjamin Young: You may be dealing with bad contexts. Other groups don't want the RDF model, they just want the JSON.
... For some people, JSON-LD is just JSON. They may not want to deal with graphs.
... Google doesn't deal with contexts, the presume that the data uses their vocabulary and interprets it that way.
... You can just look at the JSON-LD context references and use them as tokens for things you already know, rather than actually interpret it at runtime.
Sebastian Kaebisch: It is similar to the history of the Thing Description.
... We want to increase interoperability and semantics helps with this.
... But some people can't be bothered with that. JSON-LD allows us to satisfy both concerns.
... It comes down to the authors, and if they are using the tools properly.
Pierre-Antoine Champin is scribing.
Gregg Kellogg: You are coming from different perspective.
... sebastian starts with a JSON format, which JSON-LD allows to semanticize.
... rwaren is faced with bad data.
Benjamin Young: Sometimes Turtle is the best tool for the job.
... There are government agencies using "antique" RDF vocabularies. They still support antiquated formats. They may support JSON-LD, but it is often broken.
Rob Warren: The problem with saying it is tooling ...
... Serialization is JSON, but data structure is RDF.
... Libraries permit people to be lazy.
Rob Warren: Certainly, WoT wants short documents, but I don't care about that. Better to have it done right.
Benjamin Young: JSON-LD does try to bridge worlds, but you need to know what side of the bridge your on. You talk about it differently to different audiences.
Benjamin Young: People like cookbooks for thinks like Open Badges.
... Those people don't care about graphs, but may later on.
Pierre-Antoine Champin: Reminds me of some discussions where JSON-LD was discussed as a good language for JSON or RDF, but not both at the same time.
Benjamin Young: If you use the schema.org vocabulary, you can make a bad data model, if you just use Name, and City without Postal address.
... Complexity is not the problem, Complications are the problem.
Gregg Kellogg: Some best practices on publishing JSON-LD in general services might mitigate some issues.
... For example "don't use external context, use inline contexts".
... Do we consider context IRIs as URLs or as identifiers of "blocks" of context.
... Dereferencing context also raises security issues.
Benjamin Young: There are some actionably things the JSON-LD community can do. E.g., the context authoring bit can be made more explicit.
... There is an interest in vocabularies, but they're not contexts.
... They need a context tool where they can pick terms defined in other contexts.
... There are tooling pieces that are missing; if you don't read the documentation, but not necessarily how to use them.
... More tooling might help to bridge the gap between finding a context and using it to create some data.
Anatoly Scherbakov: About generating contexts, how about generating contexts from ontologies.
... If an rdfs:range of some property is a Person, we know that the value should be an ID and not a literal.
Benjamin Young: We went further at Weily, but used it for our own purposes.
... The idea was to stuff as much as could be done into the context.
... the RDFS part was hard to maintain.
Pierre-Antoine Champin: There is a JSON-LD for Prov that hasn't been published yet.
... I asked for some wording changes, as the context talks about ranges, but that's not what it does.
... You can craft JSON-LD where values don't correspond to the context expectations.
Benjamin Young: Most people want JSON-LD to be something it's not.
... They want it to be JSON Schema, or something else.
... We need to address these community expectations.
... We need some JSON-LD adjacent tooling to help visualize the results and verify that it meets your expectations.
... We're trying to make context creation easier at Digital Bazaar.
Rob Warren: I deal with people all over the complexity tree, and it's hard to talk to people that don't want to understand the specs, are find them too confusing.
Gregg Kellogg: Something similar came up in the RDF-star WG: things go wrong when your data does not match the shape that you are assuming.
Benjamin Young: One of the biggest things that's helped in JSON-LD is "safe mode". It allows you to find things that are not matching expectations.
... You must define terms to use them.
Anatoly Scherbakov: I market YAML-LD as I think this makes things generally easier. For instance, you can have comments.
... I just learned about Link-ML, which looks like a competitor to YAML-LD.
Benjamin Young: Not a competitor, other than using YAML.
... Link-ML is more about data modeling.
... It's between a vocabulary and a context file.
... Is there some thought about signing contexts? For instance by adding a hash.
Benjamin Young: There's a breakout session on this on Wednesday.
,,, Passing a hash fragment could be used by a document loader to retrieved a cached context.
... SSL only gets you so far. Document Loaders allow for better control by creating a white-list, for example.
... At one time, the value of @context could have been a value containing a hash.
... There was also the Hashlink proposal, which didn't go anywhere.
Pierre-Antoine Champin: A lot of things broke when W3C went from HTTP to HTTPS.
Rob Warren: I publish HTTP ontologies, but some people want HTTPS in front of everything. I have hacks to try to reconcile them.
Alexandre Bertails: We require people using namespaces to have them within w3.org/ns, as you can just use the name without worrying if it is http://w3.org or https://w3.org.