Print

Print


*****  To join INSNA, visit http://www.insna.org  *****

Just to add some more references, two FREE Java toolkits with good
implementations of some layout algorithms.

Jung: http://jung.sourceforge.net
Prefuse: http://prefuse.sourceforge.net

Prefuse is my favorite one. It is a beatiful piece of software well
designed and crafted. You will find there lots of layouts
(force-directed,tree,DoIT trees, Hyperbolic trees). It is fast, scalable
and with little effort you can obtain good results (have a look at the
examples).

There're lot of great ideas coming from Xerox PARC UI team and U Berkley
UI group on prefuse.

Obviously, 3000 nodes on an SVGA screen with a flat layout algorithm
(like those used in most available toolkits) is not going to provide a
good result given the density of info that  you try to graph in a
limited space. There're some ideas of using  compound graphs and zooming
based navigation to allow a smooth visualization of large scale graphs.
However, I haven't seen any algorithm implementation yet of these
compound graph algorithms in any toolkit.

Hope this will help you.

Victor Ciriza
Software Engineer
Xerox Research Centre Europe

-----Original Message-----
From: Social Networks Discussion Forum [mailto:[log in to unmask]] On
Behalf Of Valdis
Sent: 12 July 2005 00:58
To: [log in to unmask]
Subject: Re: a game of nodes and links

*****  To join INSNA, visit http://www.insna.org  *****

Shannon,

There are many common layout algorithms, but most everyone tweaks the
originals a little... a Google search should find all of the popular
ones.

I have done networks of 7,000 nodes and 30,000+ links in InFlow, but I
know that Pajek is built for large networks so they do many more...

Most layout algorithms do a pretty decent job of clustering and
structural equivalence [some focus on it more than others].  Of course
if you color or shape the nodes by various attributes you can spot other
patterns in the distribution of nodes.

BUT...

... if you work with VERY large networks you run into two problems
quickly:
1) Many networks > 2000 nodes tend to look like blobs/ hair balls ...
it is hard to see any discernible patterns
2) Your computer screen is only so large, especially if you use a
laptop.  When you zoom out to see the whole big network you loose the
ability to see colors, labels, etc.  When each node is just 1 or 4
pixels large...

The link below is to a network map of an emergent community -- all those
interested in topic X at location Y who have communicated at least Z
times.  About 1500 nodes.  Many emergent  human networks look like this
-- one large component [red], many small clusters [blue], and many
isolated quads, triads, and pairs [green].

http://www.orgnet.com/emergent_community.gif

The original graphic of the network above was almost 10,000 pixels wide.
Now imagine a network with 10x or 100x the nodes...

Valdis


On Jul 11, 2005, at 1:14 PM, Shannon Clark wrote:

> *****  To join INSNA, visit http://www.insna.org  *****
>
> Valdis,
>
> Is there a good source for those generic network layout algorithms?
>
> Also, does anyone know how large a network the common algorithms will
> scale to? (i.e. number of nodes and links)
>
> Are any of the algorithms adjustable - for example allowing for nodes
> to be arranged but also ordered/clustered by some trait? (alpha
> sorting of similar nodes, clustering of related nodes - defined in
> some manner,
> etc)
>
> Thanks!
>
> Shannon
>
> Shannon Clark
> Founder, MeshForum
> www.meshforum.org

_____________________________________________________________________
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send an
email message to [log in to unmask] containing the line UNSUBSCRIBE
SOCNET in the body of the message.

_____________________________________________________________________
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.