Re: Test of Independent Invention: RDF

On 2015-05 -03, at 03:38, Harry Halpin <> wrote:

> On Wed, Apr 29, 2015 at 3:53 AM, Melvin Carvalho
> <> wrote:
>> On 29 April 2015 at 03:11, Harry Halpin <> wrote:
>>> Not convinced. From my conversations with engineers there like Mischa
>>> Tuffield, I believe the answer is "yes" it could have been done
>>> without the Semantic Web and

I'd heard meanwhile that Garlik (pre-Experian) did benefit very much from very schema-free and being able to throw more data into the triple store at a moments notice without SQL-like schema design. Maybe we should check with Mischa.

>>> *the part of the company Experian
>>> bought*,

Did they not buy the whole company?  Pointer to that fact?

>>> i.e. the honeypot for identity fraud,  the main part of the
>>> business was done out without RDF. Thus, Experian is not maintaining
>>> the RDF infrastructure (at least 4store).
>>> So, I still haven't seen RDF used in any start-ups that have succeeded
>>> yet. I suspect there is probably some ones that *will* succeed in the
>>> healthcare space. However, in general there are major flaws in the
>>> entire Semantic Web concept ("follow your nose" URIs lead to
>>> accidental denial of service attacks,

You quote below a problem  -- a major bug -- with a Microsoft XML system, not an RDF system at all.
Or are you agains using URIs for anything.

>>> basic CS tells us graphs will
>>> always be slower than hash tables, etc.)

The first reason why that is nonsense is that graphs are hash table inside.

>>> that will likely prevent it
>>> from ever occupying the place XML or JSON has IMHO. That being said,
>>> it will likely to continue to be useful in niche markets involving
>>> data merger with dynamic schemas
>> Couldnt every statement you made above about the web of data, be applied to
>> the web of documents, and be contrary to experience?
> Melvin - which is why Google exists.

Google.  Yes, that's the company which brought its search engine into the modern age using a huge internal Knowledge Graph, which now drives much or its operations.  Yes, it does not share it in general -- but then, would you?

That's the company which by reading semantic web data in microformats and RDF/a in its crawl has prompted approaching 30% of web pages on the entire WWW to contain RDF-equivalent data?

> The reason why Semantic Web stuff
> doesn't scale in most real-world apps would be that you would
> basically need a Google-style infrastructure.

You glibly roll that off without any consideration of what sort of an app, what sort of a problem, and what sort of a scale, none of which are simple questions with one-line answers.

> Yet search over Linked
> Data seems to have stopped working (Sindice) and I haven't heard of
> real-world caching. But for a non-SemWeb example of "follow-your-nose"
> failing hard, when W3C made XML processors think the XML DTD had to be
> retrieved from,

- W3C (XML Schema)  did NOT specify for XML processors that  the DTD should be retrieved.
- That event  was only one buggy implementation which did. 
- You are talking a non-semantc web implementation anyway.

> the server basically couldn't handle this
> well-meaning DOS attack :)

> However, I do think the DOS attack
> problem/caching/searching are very solvable.

So you are a fan of follow-your-nose or not?

> Another  reason why Semantic Web stuff doesn't actually scale is basic
> computer science and so isn't likely solvable

I'd like to see an actual elaborated argument there rather than political rhetoric.

> -  and the reason we are
> seeing JSON take off (rather than RDF) as the lingua franca of the
> Web: array-values pairs map well to hash tables and what programming
> languages actually do. I would be shocked if graph DBs (see travelling
> salesman problem) ever got nearly as fast as hash tables (O(1) vs NP
> complete), so thus in general I think as a core technology the main
> problem with moving to RDF is a huge performance loss.

You are confusing two things.  It is true that JSON is appealing because it is a native data structure to the current functional language of the day.  That is natural.  It also is superior to XML for hierarchical data in that it has numbers as well as strings.  It is inferior to (say) turtle in that it has the JS problem of only having one number type, whereas in turtle 2, 2.0 and 2.0e0 are distinct typed numbers which are generally expected in the data world, python, etc etc.

Then you are claiming RDF is inferior because graph problems are in general harder to solve than tree problems.  This is extremely disingenuous. The traveling salesman problem on an arbitrary graph is hard to solve no matter what data format and model you use.  It is just going to be easier to code using a language which handles graphs.   A tree-like query, on the other hand, will be fast ether you write the tree in JSON or Turtle.   

A triple store is just hashes inside. 
Yes, you pay a bit of penalty for having the extra possibilities but only a bit.

> Also, it would be useful if Semantic Web people really thought through
> decentralization. URIs are not decentralized, they are rented from
> ICANN, which runs a number of quite centralized name-servers. Yes,
> once you buy one you can mint infinite URIs, but that's pretty far
> from decentralized - and TimBL has said as much: "We could
> decentralize everything but this"

Please don't  quote me out of context.

Note I am really involved in a lot of work on re-decentralizing the web which does use URIs with domain names in. How so?    Because the value of that is massively more than then damage currently inflicted by the DNS.  You mint a new URI with every loop in a program. You only actually need to create a new domain name rarely, such as once in a project.  Yes, the philosophical basis of the naming, or the commercial arrangements may not be perfect but that argument is a million miles beneath that of the benefits of RDF.

You go away and replace DNS with something sounder politically and commercially and RDF will use it straight away of course.   But keep your campaigns against RDF and DNS separate.

> That being said, I agree with Juan - in specialized cases involving
> data merger and a natural graph structure, Linked Data makes tons of
> sense. I think the domain of health care is likely to work out in real
> companies, and likely social network analysis for the
> military-industrial-surveillance complex. Can't think of too many
> other domains where it makes tons of sense off the top of my head, but
> would be happy to hear more and hope to see many SemWeb related
> start-ups make the next million bucks.

Well, if they see the encouragement you have given them in this thread, they will probably roll over and die, but I note your faint praise for them.

>   cheers,
>      harry
>>> And as a source of academic papers :)
>>> On Tue, Apr 28, 2015 at 8:58 PM, Bob DuCharme <> wrote:
>>>> I never said that they were purchased "due to RDF." Sampo asked about "a
>>>> company or consortium out there which has made 1-10 million bucks
>>>> applying
>>>> technology, which couldn't have been without the Semantic Web." Garlik
>>>> applied this technology and made a million bucks, so they were an
>>>> obvious
>>>> answer to Sampo's question.
>>>> Could they have done it without RDF technology? See what their CTO Steve
>>>> Harris said at
>>>> Bob
>>>> On 4/28/2015 5:51 PM, Harry Halpin wrote:
>>>> On Apr 28, 2015 9:59 AM, "Bob DuCharme" <> wrote:
>>>>> On 4/27/2015 5:08 PM, Sampo Syreeni wrote:
>>>>>> All of this Semantic Web stuff has existed for a while now. One would
>>>>>> expect that there is a company or consortium out there which has made
>>>>>> 1-10
>>>>>> million bucks applying technology, which couldn't have been without
>>>>>> the
>>>>>> Semantic Web.
>>>>> If you're looking for a dramatic success story in which one company is
>>>>> 100% about semantic web technology and then makes a million dollars,
>>>>> here's
>>>>> one:
>>>> Bob, they were not purchased due to RDF. Their triplestore and use of
>>>> RDF
>>>> was at best support for their main project  They were purchased because
>>>> they
>>>> would use honeypots to identify identity fraud. It's possible they used
>>>> RDF
>>>> to help combat identity fraud, but they were not purchased because of
>>>> RDF.
>>>> That's like saying a social networking company was purchased because
>>>> they
>>>> were using this thing called a SQL database :)
>>>> That being said, there's more investment in RDF than there used to be.
>>>> Has
>>>> the technology hit a home-run like XML and taken over the industry?
>>>> The honest answer is "no, not yet." And XML is rapidly being eroded by
>>>> JSON
>>>> and Javascript. Who knows what will be next?
>>>>   cheers,
>>>>         harry
>>>>> Companies such as TopQuadrant, Franz, and Cambridge Semantics are doing
>>>>> just fine, and more importantly, their customers are doing quite well
>>>>> using
>>>>> this technology. I think the more interesting thing to look at is the
>>>>> number
>>>>> of well-known companies that while not devoting themselves 100% to this
>>>>> technology, are still getting more and more work done with it:
>>>>> It's been interesting to see different divisions of Bloomberg joining
>>>>> these ranks lately.
>>>>> Bob DuCharme
>>>>> @bobdc

Received on Wednesday, 6 May 2015 06:27:01 UTC