[rdfweb-dev] Re: Parsing FOAF in perl?

[Danny Ayers]

>Okay, I've had a shot at adding a SOM to foafnaut, I found an algorithm
>(this was before this recent discussion) in
>http://members.ozemail.com.au/~dekker/snavis.pdf and have tried to
>implement it, but it doesn't seem to work brilliantly, there's probably a
>bug somewhere in my implementation The main problem is people all an
>equal distance from a person normally seem to all end up on top of each
>other * but I think the methodology is certainly feasible, and the
>performance is surprisingly good.

Looking good so far. That paper looks a great find. It's funny, I hadn't
really been thinking about the SOM stuff in layout terms (e.g. like
spring-embedded), rather as a way of (potential) autocategorisation. I'd
completely missed the obvious approach. I've not read the paper yet, but am
very curious what variables are you plugging into the algorithm - what is
the significance of two people being close together on the map?

>http://jibbering.com/2002/12/foafnaut.png is an example of what you get,
>the clustered people are all clustered on their nodes. Unfortunately it
>often seems to crash my Adobe SVG with a "Pure virtual function call
>error" which doesn't stop the script continuing, but you can't use the
>browser after you dimiss the error.

I don't get the error (IE6, fairly recent patch, Win2k).

>Implementing this means I've had to implement Damians path finding
>algorithm, which is very useful, although how paths might be exposed in
>the foafnaut interface is an "opportunity", especially with the stability
>problems making it difficult to have large datasets.

Have you got a ref for Damien's algorithm please.

>if you want to try it out yourself,
>http://jibbering.com/foaf/foafnaut.som.svg open up a load of people, then
>click on the key button (we've still not got a key, so for now I've taken
>it over.)
>
>Any feedback useful.

Apart from the Rugby Player (heh) effect, the only bit that stands out is
the jumping around of the whole lot in the early cycles. I've got it running
with Edd highlighted, it'd be good for him to stay centre screen - I wonder
if the adjustments could be made without too much overhead.

I'm really pleased this Kohonen stuff has resurfaced in this context - I
reckon it's fascinating & potentially dead useful. One idea - imagine that a
system scrapes the content from the pages included in peoples foaf data.
Generate a vector of values from word frequencies, and use that as input to
the SOM. People with similar pages (and hence presumably interests etc)
should get grouped together.

Cheers,
Danny.