> reliance on the web infrastructure (IRI, DNS, web servers, ...)
This is the whole point of RDF. It is what enables Linked Data.
> I read the history of JSON-LD by one of its main author [1], and it does not originated in the RDF community.
Not sure what you mean exactly, but here’s from http://manu.sporny.org/2014/json-ld-origins/:
> JSON-LD started around late 2008 as the work on RDFa 1.0 was wrapping up. We were under pressure [...] to come up with a good way of programming against RDFa data.
JSON-LD definitely was always an RDF serialization, created by people intimately familiar with RDF and the Semantic Web.
Exactly. That's my exact point. So, now, my problem is other scholars asking "why don't you use our model X or Y based of RDF" when the only thing I care about is the graph part. And I don't use RDF because of the points I exposed earlier. I should I've put emphasis that I'm working in a given field (lexicography), and that the RDF issues I face may be irrelevant for some other use cases.
As for the JSON-LD origin I stand correct.
Also, nodes can host complex structure but do not expose it to the graph (edges are made only between nodes) and edge can be of n-arity. The system is regular enough to implement simple and complex lexicographic constructs easily as well as to use the graph in an actual application.
Issue with RDF et al if that before you start even working on your real data, you have to come up with an ontology, that is to describe the world in categories, which is not easy. Another problem is that a lot of projects want to define standards that works with everything. That's not realistic. So I took an approach where you can learn from your domain by working directly on your data, make something useful with it, and then expand it you need it. KISS.
I know that some of my collegues are using JSON-LD as the interchange format for this kind of thing.
Prolog won't come with reasoners for the description logic profiles supported by OWL 2 (EL, RL), but can be used to implement such reasoners if needed.
Imho, JSON-LD is a piece of crap (because it is based on JSON, that is flawed by design). So I would definitely promote the usage of RDF/N3 (and its N3.parse()) as the way to go when you need to transmit typed graphs from server to client.
Note: N3 is a dialect of RDF, that is super easy to serialize and deserialize, that builds graphs of objects in memory (which JSON cannot do, because it can only describe trees). And that (almost always) follows a normalized vocabulary for types, IDs and labels.
(as you can see, nothing very complex :)
{ [ x:firstname "Ora" ] dc:wrote [ dc:title "Moby Dick" ] } a n3:falsehood .
PS: yep, there are blank nodes here. PPS: very few parsers take this into account, I agree. But I am pretty sure N3.parse() does.
> JSON-LD is trying to make RDF sexier (more exactly said, RDF try to pimp itself by using JSON-LD) but as it is just RDF with a new serialization syntax, it share every of its issues.
JSON-LD is _only_ a serialization format. Look, here are _nine_ others. https://help.poolparty.biz/doc/developer-guide/general-infor... Choose the one that best suits your needs. And using language like sexier and pimp, words fail me.
> a literal node (strings which carry the useful information for an human) cannot be the source of an edge, so you end up creating "resources" nodes and pushing content as leaf of the graph. Access to its is thus done with a lot of indirections.
What you mean here is that in the [S, P, O] resource, where S is the subject and P is the predicate and O is the object, one can't make the Subject a string literal. No you can't. Big deal. Whatever language and language framework you're using should have an abstraction for unpacking string literals, and if it doesn't you can roll your own. The semantics of RDF are that literal values are Objects. This is because in natural language we tend to say things like [User5000, is called, "titanix2"] and not ["titanix2", is the name of, User5000]. Sorry if you don't like it, that's the convention: https://www.w3.org/TR/rdf11-concepts/#section-Graph-Literal
> no annotations, which mean you cannot represent simple graph construct such as a weighted edge.
No idea what this means
> blank nodes, needed to join more than two piece of data which, are anonymous so they are difficult to reuse. That's sad because a graph is supposed to make data link each other. Coupled with issue 1, it mean most linguistic resources graphs are actually closer to trees than real graphs.
I didn't get how blank nodes worked originally and made myself look like a fool when I questioned how they worked on a mailing list. Such is life. They actually make sense for what they do once you wrap your head around them. Rather than me explaining them to you here I urge you consider how they work again: https://www.w3.org/TR/rdf11-concepts/#section-blank-nodes
> reliance on the web infrastructure […]
Well, it's not called the semantic web for nothing! Also, it's been pointed out to me that one can use URNs instead of URLs if you don't want to use the semantics of the HTTP schema. But you know what it seems to be working for pretty much everybody else and how expensive is it to but a domain these days?
> performances is abyssal. I tried one of the dictionary build on RDF, and some query took up to 40 seconds (!) to return no result. Even bloated web dictionaries perform under 2 seconds.
Without further details I'd have to presume you're talking about SPARQL? Yes some queries are slow but if you're dealing with millions or billions of triples then that's inevitable. I've discovered that you can optimize SPARQL once you start digging into it. No it's not easy, but it's not exactly simple tech either. Without details about the triple-store you were querying against and the query you used there's nothing more I can help you with. Anyway, your complaint is like you having a slow query on some particular relational database and you blame the entirety of SQL, wow.
For annotations, you can read[1]. What I meant however a really simple graph construct: information on edges. In the Neo4j lingo that is called a property[2]. There absence in RDF make it difficult to represent situations such "Paris"---distance:54---"Rennes". I'm visibly not the only one it's itching[5].
Yes you can use URN, be some projects doesn't support them. Check out the linguistics data cloud[3] for an example with such a requirement. Moreover, renting a domain (you can't buy a domain name) may be cheap but why putting this burden on yourself if you can avoid it? Even Microsoft mess up with them[4] for time to time. I read some papers of 2 years old and all the endpoints and links where dead. So yes, if you're not doing a web project, no need for a web format. Sadly this is what is done in my field.
Anyway I'll be curious to see something you built with RDF, if any. I failed to find your work besides two papers that aren't related to the subject.
[1] http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.352...
[2] https://graphaware.com/neo4j/2013/10/24/neo4j-qualifying-rel...
[3] http://linguistic-lod.org/
[4] https://www.theregister.co.uk/2003/11/06/microsoft_forgets_t...
So create your own serialization format suited to your own needs.
> I don't really understand the point you try to make about literals. Mine is that because literals cannot be the origin of an edge, they form sinks. And that's very annoying for building a graph (of strings) because you then need to create proxy nodes for data all the way up.
If by origin you mean what RDF calls subject then yes, as I said, the RDF data model does not support subject literals. That is by design. What you could do is create a resource that encapsulates strings in the way you want a use that for both the subject and object. Don't think them as proxy nodes, think of it like you now have the ability decorate your string literals in an extensible way.
> For annotations, you can read[1]. What I meant however a really simple graph construct: information on edges. In the Neo4j lingo that is called a property[2]. There absence in RDF make it difficult to represent situations such "Paris"---distance:54---"Rennes".
PREFIX distance: <http://www.my_cool_domain.fr/prop/distance>
[wdt:Q90 distance:354 wdt:Q647]
> I'm visibly not the only one it's itching[5].
Two wrongs don't make a right. Considering that link was calling for this in 2010 suggest to me that you oughtn't hold your breath. Also them self-describing as an expert is crass.
> Yes you can use URN, be some projects doesn't support them.
Well, that is not the fault of RDF now is it?
> Moreover, renting a domain (you can't buy a domain name) may be cheap but why putting this burden on yourself if you can avoid it?
If you can't afford €10 per year or the institution where your project is located can't afford it then you have bigger problems than imaginary RDF restrictions.
> Anyway I'll be curious to see something you built with RDF, if any. I failed to find your work besides two papers that aren't related to the subject.
I've used Wikidata and DBpedia to build a corpus of philosophers and their philosophical texts without too much of a problem once I wrapped my head around how RDF and SPARQL work.
Paris on Wikidata: https://www.wikidata.org/wiki/Q90
Rennes on Wikidata: https://www.wikidata.org/wiki/Q647
[1] https://www.safaribooksonline.com/library/view/practical-rdf...
Main issues are:
- a literal node (strings which carry the useful information for an human) cannot be the source of an edge, so you end up creating "resources" nodes and pushing content as leaf of the graph. Access to its is thus done with a lot of indirections.
- no annotations, which mean you cannot represent simple graph construct such as a weighted edge.
- blank nodes, needed to join more than two piece of data which, are anonymous so they are difficult to reuse. That's sad because a graph is supposed to make data link each other. Coupled with issue 1, it mean most linguistic resources graphs are actually closer to trees than real graphs.
- reliance on the web infrastructure (IRI, DNS, web servers, ...) which is a big dependency with a lot of gotchas and not fitted for every uses cases (self contained data for example)
- performances is abyssal. I tried one of the dictionary build on RDF, and some query took up to 40 seconds (!) to return no result. Even bloated web dictionaries perform under 2 seconds.
Also, the second sentence of the article is totally false. I read the history of JSON-LD by one of its main author [1], and it does not originated in the RDF community. It originated elsewhere, and then there was RDF people included in the loop.
[1] http://manu.sporny.org/2014/json-ld-origins-2/