W3C home > Mailing lists > Public > semantic-web@w3.org > August 2012

Re: RDF2RDB 0.5

From: Paola Di Maio <paola.dimaio@gmail.com>
Date: Mon, 6 Aug 2012 12:09:44 +0100
Message-ID: <CAMXe=SrzvmWhMJnuuXod2GF-YZK6+-mSs9Lu0kdBTHG+=PYKtg@mail.gmail.com>
To: Stephen Williams <sdw@lig.net>
Cc: semantic-web@w3.org

thank you for thoughtful reply

I do not want to defend the relational model per se, which by all
means has its limitations (does not supPort semantics was one of its
main criticism at the time from what I recall)

Of course I totally agree that unstructured data (such as what
contained in natural language) is semantically richer and potentially
more meaningful

However , from what I know, only data that is modeled according to the
relational principles can produce reliable, scalable output with
'integrity'.. If you model data according to the relational
principles,   mapping RDF  to DB /sql is trivial

Can RDF resources on the open web be queried meaningfully (and
economically in terms of resource deployment) and produce reliable

fromwhat I gather , only 'in principle'and up to a point, ie, in a
controlled environment

until we cannot obtain reliable output of queries on the SW,
we can forget about achieving reasoning capabilities-

to move to the intelligent web, we need data integrity and
reliability, and as far as I know only the relational model can
guarantee that, but I may be wrong. Open to be shown otherwise.
(maybe you know something I dont know?)

One of the things that surprises is a Semantic Web that can be only
queried from dumps (a la DBpedia), I agree it may be great work, but
its not semantic web as i figure it

anyway going back to the original question,

the way I see it a 'reasoner' capable of supporting the following
any schema>>>  transform to >>>any other schema
say, rdf data>>>> transform to >>> SQL

could easily be developed with rules

but the problem i was trying to highlight is that scriptwise
rdf to sql is  trivial,  however data integrity/data model wise,
the transform is not trivial at all

one can have DB data that spits false results (if not properly
modelled), that is the worst that can happen and the great risk

a reasoner  or decision support system (as a web service) capable of automating
the structuring of data (so that it can be queried with integrity)
would be ever so useful and imho, potentially groundbreaking

but maybe in the RuleML community could share some enthusiam to
explore the conjecture  :-)



On Mon, Aug 6, 2012 at 7:45 AM, Stephen Williams <sdw@lig.net> wrote:
> On 8/5/12 7:52 AM, Paola Di Maio wrote:
> could be very useful!
> The tool is pretty cool and, from the paper, looks useful.  The mapping of
> property URI to property table name will be problematic in some situations.
> The number of tables that would result is another potential issue.  Still,
> if you have a powerful relational environment, could be very useful.
> What is interesting to me is that the exact same logic that RDF2RDB uses
> could be used to provide equivalent SQL capabilities over a triple / ntuple
> store while still being an ntuple store.  Things like property tables could
> be, one way or another, views into subsets of triples.
> (anything that avoids the use of sparql must be useful)
> </joke>
>  however, suspect that there is a modelling challenge
> and possibly some software already exists that does that
> modelling challenge:
> to create relational databases,  it is necessary to model data
> ie to define  what to put in columns , in raws, the granularity, and a
> bunch of other stuff, including normalisation (check codds relational
> principles)
> Part of the point of RDF / Semantic Web technologies is to avoid the need to
> do this kind of thing, at least in the same sense.  Once you can clearly
> express meaning in a stable, distinct way, it shouldn't be necessary to do
> this kind of work to decide what form it is in for particular applications.
> I think it is more correct, in an RDF/Semantic aware context, that to create
> a relational model:  You have to decide what to data elements to group and
> to think about all of the queries you might want to do in the future.  You
> then have to create columns that can be used to do those queries.  If you
> are starting with RDF/semantic data, you then throw away all of the
> relationships in the data because that is modeled implicitly in a normal /
> normal form relational model.  Or the relationships all become their own
> tables, as this tool models it.  Relational models take more human effort
> because you have to anticipate how they are going to be used to know how to
> arrange data and what to model implicitly, through join queries etc.  When
> you only have a few data objects that you bother to put into a database,
> this seems reasonable.  It is a lot less reasonable when you have a lot of
> variability and start thinking of representing everything in a database /
> knowledge base.
> You can model triples in relational tables as triples, as Oracle and others
> do [1], but that is explicitly not a normal / traditional relational model.
> [1] http://download.oracle.com/otndocs/tech/semantic_web/pdf/rdfrm.pdf
> this is conceptual work that only an intelligent and trained human can
> do (that I know of) and even so, its hardly 'right' or 'wrong' rather
> a function of purpose (variable, not standardizable)
> RDF, from what I know, is typically unstructured data
> (unless it has been RDFIZED form data previously modelled and
> extracted from DBs, but I assume this is not the kind of RDF that you
> are targeting here)
> Of course you can use either structure to contain any kind of data or data
> structure one way or another.  In the way they are meant to be used:
> I would never think of RDF data as being less structured than relational
> data.  It is the opposite: You can easily model all of the data in a
> relational model, plus explicitly represent all of the relationships between
> data plus use a system of globally shared semantic concept URIs.  You could
> say that RDF has a flexible, not narrowly fixed and predefined structure.
>  what you propose, as I understand it, could be useful if you can get
> the system to model/represent rdf unstructred into
> meaningful/structured relational data
> If however you could come up with a web service that enables a user to
> create structured data models from unstructured ones, then I think
> that would be groundbreaking work and I d love to see something
> (unless its already available and I am not aware of it then please point)
> In the long term, there's little point in putting data in relational tables
> except A) to implement a triple store / semantic model underneath or B) to
> interface with legacy systems or software that operate only on relational
> queries.  Everything that you can do with SQL could be done over an
> RDF/Semantic model, plus a lot more.  Any existing performance shortfall on
> the triplestore / SparQL side can be fixed, especially if you just want to
> query over relational-level data (highly homogenous structure).
> cheers
> On Sun, Aug 5, 2012 at 2:03 AM, Michael Brunnbauer <brunni@netestate.de>
> wrote:
> hi all
> to support the proliferation of RDF, I have begun work on a tool to convert
> RDF data into relational databases. It is intended for people who want to
> use
> data available only as RDF but do not have the patience to install a triple
> store and learn SPARQL. I know there are tools that present a SQL view
> over a triple store but installing such an environment also is quite
> challenging.
> There has been work on real "converters" in the past, especially the paper
> from
> Wajee Teswanich and Suphamit Chittayasothorn:
> http://hcotuk.etu.edu.tr/semanticweb/A%20Transformation%20from%20RDF%20Documents%20and%20Schemas%20to%20Relational%20Databases.pdf
> My tool works much like the one described in this paper (has it ever been
> implemented ?), with important differences:
> -It supports incremental runs (adding more RDF data to a database later)
> -It supports entailment (rdfs:subClassOf, rdfs:subPropertyOf, rdfs:domain,
>  rdfs:range, owl:equivalentClass, owl:equivalentProperty,
>  owl:FunctionalProperty)
> The software is written in Python and uses the RDFLib Python library. It
> works
> with local and remote files. You can find more information and download it
> at
>  http://www.netestate.de/De/Loesungen/RDF2RDB
> There also example database dumps for
>  http://www.w3.org/People/Berners-Lee/card.rdf
> and
>  http://www.w3.org/People/Berners-Lee/card.rdf
>  +
>  http://xmlns.com/foaf/spec/index.rdf
> there.
> Don't rant about the invalid skolem URIs - you can set the skolem URI prefix
> to something valid in the configuration :-)
> So far this has just been fun and I would like to have some feedback on the
> usefulness before I spend more time on it.
> Regards,
> Michael Brunnbauer
> --
> ++  Michael Brunnbauer
> ++  netEstate GmbH
> ++  Geisenhausener Straße 11a
> ++  81379 München
> ++  Tel +49 89 32 19 77 80
> ++  Fax +49 89 32 19 77 89
> ++  E-Mail brunni@netestate.de
> ++  http://www.netestate.de/
> ++
> ++  Sitz: München, HRB Nr.142452 (Handelsregister B München)
> ++  USt-IdNr. DE221033342
> ++  Geschäftsführer: Michael Brunnbauer, Franz Brunnbauer
> ++  Prokurist: Dipl. Kfm. (Univ.) Markus Hendel
> Stephen
> --
> Stephen D. Williams sdw@lig.net stephendwilliams@gmail.com LinkedIn:
> http://sdw.st/in
> V:650-450-UNIX (8649) V:866.SDW.UNIX V:703.371.9362 F:703.995.0407
> AIM:sdw Skype:StephenDWilliams Yahoo:sdwlignet Resume: http://sdw.st/gres
> Personal: http://sdw.st facebook.com/sdwlig twitter.com/scienteer
Received on Monday, 6 August 2012 11:10:15 UTC

This archive was generated by hypermail 2.4.0 : Tuesday, 5 July 2022 08:45:30 UTC