Re: DBpedia as possible project

Peter,

On Mon, 2010-11-01 at 17:43 -0700, Research wrote:
> I agree with Murray's concerns about (all-too-common) modelling abuses: for
> example (and I'm sorry for picking on Lars and this example but it's a good
> bad one), an astronaut is not a specialised class of person - it is a role
> type played by a person type - and when badly modelled, leads to screw-ups
> and false inferences later on: what happens when the person is doing
> something other than being an astronaut (being a Playmate for example?)? or
> is no longer an astronaut? Roles help give context to people's relationship
> with others and the world.
> 

The model in question wasn't from Lars but from DBpedia, which Lars has
already commented needs a lot of work.

Certainly the modeling I saw was poor to say the least but it is a
starting point and certainly topic maps would be one way to improve it. 

> I disgress, but Topic Maps explicitly models roles in a way that other
> standards do not - and the topic map community really needs to play to its
> own strengths - such as the quality of its underlying model and intellectual
> rigour in weeding out poor modelling approaches.
> 

Well..., but you can poorly model with topic maps just as you can with
any other technology. 

But I take your point to be that we should develop good models of topic
mapping for others to follow. 

> There have been several comments over the last couple of weeks about the
> "added value" of topic-mapping" this or that and my 0.01 worth would be to
> suggest that we pick a project/domain/hobby that can:
> - highlight, make use of and enforce good modelling practices (such as
> tackling the above example);
> - allow valuable searches and deductions to be made on datasets from better
> modelled information territories as a result.
> 

What do you think about the suggestion that we work on DBpedia? 

> I picked on the role example above because, in every discussion that I have
> had with businesses and government agencies over the years, this issue has
> come up, largely because of dB admins trying to force square pegs in round
> holes and (ab)using class specializations as a way to tackle, for example,
> job and functional roles, relationships, etc. There is consistently an "aha"
> moment - people get it. And we should be agents to move that sort of
> understanding forward.
> 
> It might be an interesting thread on this forum to identify what could be
> the "unique selling point" of TM vs other standards and identify *from that*
> what would be a suitable starting point...

It probably would be an interesting thread but I think I would prefer to
see the unique virtues of topic mapping emerge out of something like a
DBpedia project.

Hope you are having a great evening!

Patrick



> 
> All the best,
> Peter
> 
> ---------------------------------------------------------------------
> Peter F Brown
> Independent Consultant
> P.O. Box 49719, Los Angeles, CA 90049, USA
> Tel: +1.310.694.2278
> 
> Transforming Business with technology, Standards and Interoperability
> www.peterfbrown.com
> ---------------------------------------------------------------------
> 
> -----Original Message-----
> From: topicmapmail-bounces-Zo64W7twoUFWk0Htik3J/[email protected]
> [mailto:topicmapmail-bounces-Zo64W7twoUFWk0Htik3J/[email protected]] On Behalf Of Murray Altheim
> Sent: Monday, 01 November, 2010 15:16
> To: topicmapmail-Zo64W7twoUFWk0Htik3J/[email protected]
> Subject: Re: [topicmapmail] DBpedia as possible project
> 
> On 02/11/10 10:14, Patrick Durusau wrote:
> > Lars,
> >
> > I decided to at least poke around in the top level classes of DBpedia,
> > at: http://mappings.dbpedia.org/server/ontology/classes
> >
> > This could be fun.
> >
> > Did you know that PlayboyPlaymate is a top level class? Right next to 
> > Philosopher? Who knew?
> >
> > BTW, if you look at "Properties on PlayboyPlaymate:"
> >
> > You find what you expect, bustSize, hipSize, waistSize, but also:
> >
> > anthem?, boiling point?, dateOfAbandonment?, highestPosition?, 
> > lowestPosition?, militaryBranch? (is that a preference?) 
> > mouthPlace?..., and it just goes on.
> >
> > When I got to timeInSpace, I realized that not every property was 
> > probably filled in for a PlayboyPlaymate.
> 
> Hi Patrick,
> 
> The project that I described in Montreal at Balisage 2008 was essentially
> DBPedia implemented in Topic Maps over JSPWiki. It generated an XTM
> snapshots of the current wiki on demand as a web service. That project was
> sadly killed by a new director who thereafter resigned (he was not
> particularly effective), but there were some lessons learned.
> 
> One of the things I find disturbing about the quick jump from source
> availability to modeling is the very typical disregard for semantics in
> favour of inferring suitability based on a term or predicate's label, often
> without even bothering to read the documentation (when that even exists).
> This is exemplified in the common misuse of 'equals', 'same as',
> 'equivalence', 'identity', mixing and matching concepts from different and
> often incompatible sources, mistaking human statements for logical
> relations, opinions for facts, category errors, etc.
> I could write a book about these errors; they are legion.
> Reading about zip codes and such only reminds me of this. E.g., has anyone
> once mentioned the need for something akin to a "topographical ontology", a
> "political boundaries ontology", or a "temporal ontology" as the basis for
> making assertions about politically-determined territories that change over
> time? No. That's great for experiments, but we're still operating in the
> realm of toys.
> 
> I would hardly expect that a resource such as Wikipedia, created and edited
> over a long period of time by an international community of non-experts
> using no controlled vocabularies (in mind of the actual definition of that
> term) would yield reasonable triples (using 'reasonable' in any formal
> sense). That DBPedia makes some very strange inferences about Playboy
> bunnies is hardly surprising in this environment. Even at the very beginning
> its definition of "class" is problematic.
> 
> The most important part of my cancelled project was not the mapping of an
> existing content base into a wiki capable of being harvested as a Topic Map
> (that would be analogous to the current flap over the subject that started
> this thread), it was the attempt at merging a formal library classification
> system with the user-generated informal statements on the wiki -- made
> without regard to any controlled vocabulary (e.g., almost everything on
> Wikipedia/DBPedia). That was also the basic subject of my doctoral research.
> 
> To adopt Lars' choice of words, I would submit that the use of an
> uncontrolled, "informal ontology" for purposes of machine reasoning would be
> almost completely useless, and potentially harmful.
> Entertaining, perhaps, but very poorly suited except as a hint to humans as
> to what *might* be true: any inference and interpretation of statements
> would still require human intervention. I.e., one might use this for Playboy
> bunnies but not for air traffic control (or NASA space flights).
> 
> Murray
> 
> ...........................................................................
> Murray Altheim <murray10 at altheim dot com>                       ===  = =
> http://www.altheim.com/murray/                                     = =  ===
> SGML Grease Monkey, Banjo Player, Wantanabe Zen Monk               = =  = =
> 
>        Boundless wind and moon - the eye within eyes,
>        Inexhaustible heaven and earth - the light beyond light,
>        The willow dark, the flower bright - ten thousand houses,
>        Knock at any door - there's one who will respond.
>                                        -- The Blue Cliff Record
> _______________________________________________
> topicmapmail mailing list
> topicmapmail-Zo64W7twoUFWk0Htik3J/[email protected]
> http://www.infoloom.com/mailman/listinfo/topicmapmail
> _______________________________________________
> topicmapmail mailing list
> topicmapmail-Zo64W7twoUFWk0Htik3J/[email protected]
> http://www.infoloom.com/mailman/listinfo/topicmapmail