[ba-ohs-talk] spring and force directed graphs
This is somewhat related, since we've spent some time talking about
graph visualizations. (01)
I'm currently in a small graduate seminar on information
visualization. It's a hybrid of playing with Java based tools and
learning about perception and usability. Our final project is to work
with faculty on either an interactive visualization project or
improving the existing toolkit. (02)
To my perception the toolkit is so broken at the moment that using it
for a real project is pointless so I'm choosing the route of improving
the existing toolkit. (03)
There is a current implementation of the Spring algorithm for
displaying document similarity matrices and other networks. It's not
very featured, has a poor data representation, and is inefficient. It
is being used to represent the similarity between documents discovered
by latent semantic analysis. It only works for about 100 nodes or
less. (04)
I'm going to take on the task of making it better. Given the focus
made on Touchgraph here, I'm assuming that the knowledge I gain will
be helpful in my interactions with this group. (05)
The constraints of this project require that I start from the
existing codebase and not, instead, choose to extend something like
Touchgraph. Unfortunately. (06)
However I do think there is a lot of information that can be shared
back and forth and that will be productive. (07)
Can people point me to information that has been helpful in their own
investigations of designing, programming and using algorithms for
drawing graphs? I'm especially interested in: (08)
- overviews of the reasoning used in designing these sorts of
algorithms
- design pointers for internal and external data representations of
graphs. The current system does use XML for input, but the format is
basically a wrapper around a matrix so does not represent node
interconnections easily, nor cleany allow for additions of new
nodes, or new attributes on nodes. I've seen and liked what
Touchgraph for the Link Browser uses.
- transmission formats
- [important stuff I'm not thinking about at the moment] (09)
What I'd like to have at the end of this is a system that allows for
the display of similarity matrices and clustering in the same display.
Clusters are displayed in the network and when selected can be zoomed
for detail of their contents. (010)
This is more math than I know at the moment, but I guess I'll be
picking it up. (011)
Thanks for input. (012)
--
Chris Dent <cdent@burningchrome.com> http://www.burningchrome.com/~cdent/ (013)