Re: reworking welkin

From: Arvind Venkataramani <arvind_at_cc.gatech.edu>
Date: Fri, 03 Feb 2006 19:47:25 -0500

Stefano Mazzocchi wrote:
<snip>
>> 1. What are you people using welkin for? What kind of data?
> I use it for a few things:
>
> 1) make sure that the RDF I'm writing by hand (mostly ontologies)
> make sense and all the things that should be linked are linked.
>
> 2) explore a subgraph of a bigger graph when I don't know what it is
> (normally when I need to draw equivalences)
>
> 3) explore a subgraph to understand graph-theoretical properties of
> it (how connected it is, how many 'clusters' there are) and so on
Ok. But what *kind* of data? :) I mean, do all ontologies look the same?
If you had an ontology of car specifications that needed to be presented
to a potential buyer, or perhaps a salesman, would it be different from,
say, an ontology of Java Imaging packages a programmer might use? Would
you want to do different things each, and consequently, does that have
implications for the interface design? Unfortunately, I'm not totally
sure what aspects of data I'm trying to look at, so if you could send me
sample data I'd really appreciate it.

<snip>
>> 3. What problems do you have with welkin as it is now? What are
>> the things you wish you could do, but can’t?
>
> well, first of all, it's slow. The many-body calculation algorithm is
> a greedy O(n*n) while there is an algorithm that is O(n*log(n)). That
> would give us *tremendous* speed improvement on larger graphs (up to
> 20k nodes, I would predict)
>
> second (and Paolo has been working on this but nothing has been
> committed yet, Paolo?) different node types should have different
> shapes (and not only colors).
>
>
> third, the "distribution queries" are to be rethought, mostly because
> they are not joined, but moving one query on one distribution, resets
> the query on the other one.
Hmm. How hard - since you're the guys that built this stuff - do you
think it would be to build a basic RDF inferencing mechanism into
welkin, or is there something that can be plugged in? I'd love to be
able to query on statements, but I'm not sure that can be done without
some sort of inferencing mechanism running in the backend.
>> Thanks a ton in advance, and I’m really open to suggestions:
>> user-centred design and all that :)
>
> Arvind,
>
> at the end, it all works better if *YOU* do whatever *YOU* think
> welkin is missing. Scratch your own itch and all that.
Yes, of course. I *am* doing this because I need to scratch my personal
itch - I know what *I* need, in pretty specific detail at that. But it'd
be awesome if I could get information on how people are using it, so I
can be more comprehensive in my own design. It is possible that the best
thing to do is to start with domain specific data and generalise from
that, like Vineet is doing with Relo. But I don't want to give up
without trying to make a generic interface.
>
> We will *gladly* accept patches, if you come up with any. What is
> useful for you I'm sure will be useful for others, so don't ask us
> what we want to do, ask yourself, it will greatly simplify achieving
> something with this effort that is gratifying and feels good.
>
> And don't worry if you get sidetracked and lose interest, you'll still
> be welcome here.
>
Much appreciated :)

-- arvind
Received on Sat Feb 04 2006 - 00:46:44 EST