W3C home > Mailing lists > Public > semantic-web@w3.org > October 2017

Re: [DBpedia-discussion] Semantic Web Browser

From: Adrian Gschwend <ml-ktk@netlabs.org>
Date: Wed, 11 Oct 2017 17:07:22 +0200
To: semantic-web@w3.org
Message-ID: <ea08f3a7-6d27-b31f-ad89-fbdb5a88c357@netlabs.org>
On 11.10.17 10:09, Olivier Rossel wrote:

Hi Olivier,

> I have felt absolutely no support from the Semantic Web community.

I think I know where you have been but I would like to give a feedback
as someone that is making a living on Semantic Web.

> Basically for the following reasons:
>  - very few people in the Semantic Web community actually manage
> datasets in operational conditions (so there is no linked data to
> browse, cf http://sparqles.ai.wu.ac.at/availability)

We do, as do our customers. Just because you don't see them does not
necessarily mean "very few people" are using it. In the end I don't see
any Neo4J datasets online but I don't conclude that no one is using it
just based on that fact.

>  - very few people in the Semantic Web community actually consume
> semantic data in their processes (so noone can evaluate which
> libraries/tools are lacking for a proper consumption of RDF data)

I do think that there is a problem with dumps someone once converted to
once RDF, interlinked and that was it. Unfortunately quite some projects
like that exist, many of them as outcome of former FP7 EU research
projects. While I was skeptical myself for a while I am convinced that
we now see real traction in governments and organizations moving to
production-ready linked data publication. Unfortunately we all suck a
bit in advertising them accordingly.

Regarding libraries, I preach for a while that Linked Data needs to be
better in JavaScript stacks and for that we created the RDFJS group some
time ago. The spec there is now basically done and we (Zazuko) released
a first implementation of it:

https://www.bergnet.org/2017/08/rdf-ext-v1-release/

You can find all code on Github. We do a hell lot of other useful stuff
in the JavaScript world, just browse our github repos. And yes,
documentation is again the thing we need to improve.

http://github.com/zazuko/

> But of course our point is to inspire people outside the Semantic Web community.
> And such people/companies have immediate requirements to fullfill.
> So they go the full custom HTML5+JSON way. With pretty amazing results.
We go the HTML5 + JSON-LD way. Also with amazing results :) The graph
might not be that important in the user interface (still helpful though
for many things) but you definitely want a graph-like structure in the
backe-end if you are serious about your data.

> They know RDF very well, but see no market for that.
Not sure if you heard about the famous quote about Unix: "Those who do
not understand Unix are condemned to reinvent it, poorly." I believe
this is true for RDF as well.

I am using Opendatasoft as one of our customers is using it. I pull data
that is barely usable in this form from it and transform it to proper
linked data. That is quite a task but if you want to get data in a
usable form, that's the only way to go. I see platforms like CKAN and
Opendatasoft publishing large data sets but people still have to spend
days to bring it in a form that allows them to use it in their
applications. RDF is the form they actually need and sooner or later
they will come to this conclusion.

Opendatasoft would be fare more useful if it would be built on RDF. That
is a bit a problem we have, we forget to build the tools "normal" users
can use so in that regard I'm with you.

> We must understand why.

RDF comes with a certain complexity. You can either ignore it and build
yet another proprietary silo or embrace it and start to build tooling
that facilitates creation and consumption of RDF data. We as Zazuko
chose the second option and I absolutely believe that this is the
future-oriented one. It does pay our bills and it does and will solve
real problems our customers try to crunch.

It's not that these customers did not try other stacks before, some
spent two years addressing the problem with non-RDF approaches to figure
out that this will not lead anywhere. But you will not read about those
use cases as they happen behind the firewalls.

>>From my own point of view, the success of the Semantic Web could come
> with tooling for programmer
> If we manage to provide a few things:
>  - a spec & robust implementations for rights management at named graph level

get the right triplestore and you are done. Not sure what you need a
spec for here.

>  - a spec & robust implementations for SPARQL transactions management
> at HTTP level

again get the right triplestore and you are done.

>  - a robust OGM (Object-Graph Mapper) in most major languages

What is an Object-Graph mapper?

>  - a robust REST library to auto-serialize/deserialize RDF (for ex, an
> extension to Jersey

We do this with Hydra-View, soon to be documented and renamed to Hydra-Box:

https://github.com/zazuko/hydra-view

https://github.com/zazuko/wikidata.zazuko.com

This is work in progress for a customer but what it basically allows is
to hide SPARQL queries behind a hypermedia REST API. All you do is
configure the middleware in JSON-LD and then run it. With JSON-LD
framing you can create JSON every web developer should be able to understand

BTW we also do build Web frontends that are 100% SPARQL driven, that is
not a problem at all, you just need according tooling. See
http://data.alod.ch/search/ as an example (soon to be extended to 20
million public archival records)

If you simply want to get proper JS structures from SPARQL you might for
example check out https://github.com/zazuko/d3-sparql

Again, it's all our fault that we did not build these tools before. And
yes, that would have been more useful than a "linked data browser",
whatever this means.

For providing data we did something Pubby-like, see for example:

https://github.com/zazuko/trifid

This is the back-end of all apps we build. Public samples:

* Default view: http://lod.opentransportdata.swiss/didok/8500011
* Customized for one gov entity in Switzerland:
https://ld.geo.admin.ch/boundaries/municipality/296

>  - a proper marketing of the N3.js library on the client (honestly,
> how many people even inside our community knows that fabulous lib?)

Can't say much here as we use it for a long time in rdf-ext. Also you
want to have at RDF-Ext and things like simpleRDF.

> Basically, we need a stack.
> Why not create RDFonRails, by the way :)

the future is JavaScript, like it or not. That is where it will or will
not happen with RDF. (And stores that scale but that is taken care of)

> (btw, Neo4J basically provide 90% of all that, and is pretty
> successful, so may be we should just jump on the bandwagon)

Neo4J somehow managed to convince people that they are "graph
databases". It's a mediocre product from a company which willingly
spreads FUD about RDF because they know it's the one thing that can
become a problem for them. But I have to give them the credits for
marketing, boy are they good in that.

> After that, we can again concentrate on data. (especially data inside companies)
> Honestly, noone outside the community understands (or cares) about OWL.
> RDFS+owl:domain/owl:range is enough for a awful LOT of usages.
> (once again, Neo4J provides something quite like that, and it is LOVED
> by IT developpers)

wrong (about OWL). There are people using it, every day. Just because
you do not see it does not mean it does not happen. I saw developers
with tears of joy once they enabled reasoning in the triplestore. And
why would I use something proprietary like Neo4J if I can get various
products in the RDF domain that implement a standard?

> What is important and game changers in the outside world is:
>  - typing data, and multityping it (:VERYYYYY powerful)

I don't think I get what you mean here.

>  - merging graphs coming from different sources dealing with the same
> resources for a more capable graph
That's pretty much what we do with RDF all day.

> What is extremely hard in the outside world:
>  - sharing URIs.
>  - sharing data, in general
> 
> All these points are addressed poorly by the community. Basically
> because we do not do it massively ourselves.

I have no idea where you see a problem here. RDF is the only standard to
solve these exact problems, forget Neo4J or anything alike for that. And
we do it, every day.

> But the more important advice I can give after some time spent outside
> the Semantic Web community:
> do not build a browser (you would rebuild datao.net/search.datao.net.
> Believe me, noone cares.), build a fucking awesome add-on for

on that I'm with you. BTW I do not necessarily consider myself as part
of the academic semantic web community. While I am grateful for a lot of
support I got from some of the people in there, others are not very
welcoming for outsiders like me.

> Microsoft Excel.
> 
> *That* would definitely change the way people deal with data.

I hear the excel one a lot but I'm convinced most of the time this is
un-reflected daydreaming. What exactly should such a plugin provide in
your opinion?


regards

Adrian
Received on Wednesday, 11 October 2017 15:07:56 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 11 October 2017 15:08:05 UTC