W3C home > Mailing lists > Public > semantic-web@w3.org > July 2014

Re: Statements about RDF

From: Robin Berjon <robin@berjon.com>
Date: Thu, 03 Jul 2014 15:15:41 +0200
Message-ID: <53B5577D.4000304@berjon.com>
To: Phil Archer <phila@w3.org>, semantic-web@w3.org
Hey Phil,

On 03/07/2014 13:14 , Phil Archer wrote:
> On 03/07/2014 11:20, Robin Berjon wrote:
>> My comments and predictions are not technical, they are ecological. To
>> put this differently, if it is 2.3 billion years ago and you are the
>> meanest, badassest anaerobe on Earth, you can evolve multicellularity
>> and even higher intelligence all you want, you're still going to die a
>> horrible death when the cyanobacteria oxygenate everything.
>
> By coincidence I was talking about evolution earlier this week,
> especially how a spider is never going to evolve insect eyes, or
> molluscan eyes or chordate eyes because to from its highly evolved
> spider eyes towards those would require backward steps that would be
> disadvantageous before they could move forward. Nope, not going to
> happen and nor should it.
>
> Pretty pictures of eyes in different (sub) phyla at
> http://www.w3.org/2014/Talks/0630_phila_samos/#(3)

I like your evolution metaphors, they're less nasty, brutish, and short 
than mine :)

>> The basis for my prediction stems from two observations: the vast, vast
>> majority of Web publishers I meet have no idea that RDF (still) exists,
>
> Did you have to write (still) there? Of course it F'ing well exists.

Again, I didn't write that from my point of view — of course *I* know 
RDF still exists. But it's my job: I also know about RDDL, EXI, and 
Delivery Contexts (one of which is not dead :). It's a reaction I see in 
the community relatively frequently. People remember the SemWeb buzz and 
have heard nothing since. The unmalicious "Is that still a thing?" is 
something I've heard more than once in this context.

I know it's disheartening, but it has a silver lining: not knowing 
anything about it also means no bad impressions to overcome. It's a 
whole new generation from that which tried to play with RDF for its Web 
problems 10-15 years ago and ran away screaming after seeing it had no 
notion of arrays :) I'm serious: it's a lot easier to get someone to try 
than to come back.

>> conversely I don't see any
>> movement (but I may have missed stuff) to use RDF in a way that solves a
>> sizeable tract of the problems this crowd is facing today (or that I
>> suspect will be facing soon).
>
> So we really need to look at use cases (quelle surprise :-) )

Yes, but I would be cautious about putting too much faith in use cases 
on their own.

A use case could be "I want to be able to share my data with arbitrary 
third parties in an unambiguous, interoperable way." That sounds great! 
But from that you can easily decide that in order to be unambiguous your 
terms need universal identifiers, openness to all requires fully-general 
open-world assumptions, the most general data structure to work for 
everything, etc.

Before you know it you have something that requires so much work 
*before* it shows any benefit that, like your spider eyes, it simply 
never happens.

I'd be stricter than use cases and instead look for pain. Everyday pain. 
What really, really <span lang=fr>sucks</span> in Web development today 
and is related to data?

Forms are a really good example. Managing data with forms is horrendous. 
There are libraries that hide some of the pain away, but frankly they're 
hacks. The worst part is generating forms with pre-filled data. 
Architecturally, the way validation ties into it is largely madness. You 
have to keep in sync: 1) client-side validation, 2) the form itself, 3) 
server-side validation, and (usually) 4) storage-level validation. 
Changes normally need to be made to several of those whenever your data 
evolves.

Any solution that could deliver on that and that would, almost as a 
side-effect, introduce other Linked Data properties of interest would 
IMHO stand a solid chance of making its mark.

But making it work with existing systems means that it can't be defined 
as a basic arbitrary graph in the way that the RDF DM would. A forest of 
trees though would work great IMHO.

>> Now, if I switch to a much more opinionated take I don't think that this
>> is a fatality. I do not go into details in that post, but I do think
>> that there are ways to make massive amounts of linked data emerge from
>> the Web we have, primarily by making it vernacular through solving
>> small, everyday paper-cut problems. The "Web Schema" thing I mention in
>> that post is basically the semantic web freed from the shackles of the
>> RDF data model :)
>
> Shackles? Is that how you see it? Let me create a trivial example.
>
> I take a book out of my local library which is run by a public
> administration according to a locally defined policy that conforms to
> one or more pieces of national legislation.
>
> Me
> The book
> The library
> The administration
> The policy
> The locality
> The legislation
>
> All these are actors in the chain of events and the linkages between
> them are varied. Each one gives an access point to relevant data that
> may go off in other directions too and it looks a lot more like a graph
> than a (set of) table(s). But let's make it available in a way that
> people don't necessarily have to understand the underlying graph to make
> use of the power behind it (did someone say Knowledge Graph?)

I don't think that tables have any advantages over graphs — in fact I 
find that they tend to have similar problems. For a daft JS hacker such 
as myself, you have data structures, and they can sometimes be linked to 
one another.

Data structures have boundaries: you know where they end, you can't 
navigate them and end up somewhere else. They also have structure that 
goes beyond the tabular, they're emphatically not normalised.

The difference in model (and model capabilities) between a graph and a 
linked forest are minimal, I think that in terms of modelling power they 
don't matter at all. But having a model that maps clearly to everyday 
programming constructs is IMHO a huge win. I've honestly never seen an 
RDF API that didn't make me want to hurt myself; using whatever you get 
from JSON is as straightforward as it gets. (The last time I was part of 
a project that used RDF in production, that's how it was handled. It was 
loaded at startup and converted ASAP to a JS data structure, and then we 
jumped through however many hoops it took to never have to interact with 
the RDF data again.)

Incidentally the situation you describe above feels to me a lot more 
like a forest than a graph. My arm or my name are attached to me in a 
way that a book I checked out isn't.

I also don't think that this is a huge problem for RDF. You can mostly 
constrain it a bit, and lose many of the URLs.

> For W3C Internal Use Only: I'm hoping for some sort of cross IA/INK
> Domain/DevRel thingy. Just got to work out how the heck to make it
> happen - might take a while.

I never mind an excuse to hop over the Eurostar :)


-- 
Robin Berjon - http://berjon.com/ - @robinberjon
Received on Thursday, 3 July 2014 13:16:10 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 1 March 2016 07:42:52 UTC