Re: [semweb-35] fw: Open Calais exercise
Marco Neumann
The W3C RDF validator graph might not be the easiest way to look at
the domain model and the data. BTW since you are familiar with graph
visualization engines you might recognize the visual tool that's used
here. It's based on Graphviz. There are a number of anonymous nodes
that make it more diffucult to read the graph. Your hand-drawn graph
obviously is much easier to read. Are you already familiar with
anonymous nodes in RDF?
Marco
> After absorbing the OpenCalais presentation I sat in on the other night, I
> decided to spend some time today exploring a simple, single-sentence news
> tidbit I made up. First, I graphed it by hand in PowerPoint. Then, I
> parsed the text via the Calais Document Viewer, the RDF output of which I
> then pasted into the W3C Validator in order to graph it. I placed the
> Validator's graph, my graph, and the text on a quickie web page on my site.
> You're welcome to glance at it if you like. I was at first thinking of
> starting a discussion thread for this exploration in the swnyc group and
> perhaps the swsf group, but I wasn't sure if this sort of thing was too
> trivial for the members. Basically, it could function as an example that
> would allow me to pose a lot of beginner questions to the group(s) about if
> and how statement metadata is structured by companies actually working in
> the SW field. A second potential thread might be to simply discuss opinions
> as to why my graph (what I expected Open Calais to output) did not match the
> Open Calais output and even whether the graph I expected is appropriate.
> [Alex, you are welcome to use any or all of this exploration or any
> derivative thereof in your SemanticAlley site once I feel it is cleaned up
> enough so that I don't appear too much of a dummy:-).]
>
> Here is the URL:
> http://sites.google.com/site/rickcreamer/Home/miscellaneous-items/news-rdf-exploration
>
> Rick
>
>
>
>
> --
> Please Note: If you hit "REPLY", your message will be sent to everyone on
> this mailing list (semw...@meetup.com)
> This message was sent by Richard Creamer (2to32...@gmail.com) from The
> San Francisco Semantic Web Meetup.
> To learn more about Richard Creamer, visit his/her member profile
> To unsubscribe or to update your mailing list settings, click here
>
> Meetup Support: sup...@meetup.com
> 632 Broadway, New York, NY 10012 USA
rtc1
First, let's make it clear that I am at most an advanced beginner WRT
SW. I spent several months studying, applying, and extending SW
technologies (into an Enterprise stack) about 3 years ago. However, I
have only recently become re-interested in this area and so my earlier
knowledge is stale. I am not currently using SW technologies at work
or in my non-work-related projects. Because of these, I will probably
write some dumb statements and questions.
Second, some short answers:
1) Yes, I expected the W3C RDF Validator to be verbose.
2) I was a bit shocked at how many triples came out of the NLP engine.
3) While I still need to do some more digging, I haven't yet found any
of the triples I expected amongst the output triples. I did see a few
that made sense but have not yet found any which I felt represented
core information.
4) I wasn't really trying to compare graph representations or their
efficacy for humans. Before using the Calais tool, I analyzed the
text by hand to see what triples I felt it represented. I did this as
a graph because it's a lot easier to create, think, and read when
viewed graphically, at least for me that's the case.
5) Yes, I recall anonymous nodes. I haven't really used them much and
felt, at the time I read about them, that I would prefer to find
alternative methods/patterns to represent information which would not
require anonymous nodes.
Third, this exercise was interesting to me because:
1) It allowed me to write down the data and their interrelationships
graphically without having to get bogged down in how to express them
in RDF/XML, 3-Tuples, anonymous nodes, collections, etc.
2) Even this simple example sentence brought up the need for
statements-about-statements which is one of several SW areas in which
I'm extremely interested (see last section).
3) I learned that, given the complexity of the Calais output, I would
definitely be intimidated/challenged if I had to write an application
to consume such output.
4) Verbosity, complexity, and representation aside, I was interested
in seeing how good a job Calais did in extracting meaningful
information from this little example.
5) Calais uses fuzzy metadata (an interest area of mine listed in the
below)
6) I wanted to analyze what patterns Calais used for statements-about-
statements such as fuzzy and others. I'm still working on this.
This exercise touched on the following subset of my SW interest areas:
Any sorts of "patterns" that working SW practitioners use to describe:
- Temporally changing data such has Human.JobTitle
- Statements about statements and metadata in general
(including fuzzy data)
- Configuration management (version control) for triples (ties
into temporal considerations)
- How to avoid patterns that undermine decidability
- Metadata tag sets/groups specific to certain predicates or
subject-predicate combinations
I have other SW interest areas and would be happy to discuss these
offline.
I hope this helps clarify what I was, and was not, trying to do. It
was primarily an exercise to get my hands dirty in the SW area again.
> > Rick
>
> > --
> > Please Note: If you hit "REPLY", your message will be sent to everyone on
> > This message was sent by Richard Creamer (2to32min...@gmail.com) from The
> > To learn more about Richard Creamer, visit his/her member profile
> > To unsubscribe or to update your mailing list settings, click here
>