http://img11.imageshack.us/img11/2064/testxn3.gif

And another showing start/end time.

http://img18.imageshack.us/img18/7630/testrg4.gif

Not perfect, but not bad either…16-20 is probably the max usable gradients.

1 – A simple linear scale, with time broken into n intervals and links colored based upon the interval in which they were started. Black is most recent (freshest) and lighten over time.

2 – A logarithmic scale, like the above, with greater resolution for recent events.

1/2a – A variation of either of the above where duration of the link is encoded by a gradient in the line…the originating side encoded with the start time, and the destination side encoded with the end time (up to and including black for current).

3 – A ‘node relative’ scale, with each node having links colored based upon the sequence/time that the link was established with the node relative to other links to the node. Would be screwy to look at, but might provide good sequencing information when looking at things like outbreaks. Gradients may be required here as well to properly encode the sequence.

This approach could be used for the node itself as well by coloring the border to indicate when it was introduced to the graph.

Great stuff.

The temporal coloring approach could work for showing long-term changes but you’re right it isn’t good for discerning between individual changes due to human visual “just noticable difference”. Color could be used in an animation to highlight the new additions and then have the color fade away as time marches on.

As for adding time-dimension, there are at least two things you might try. Why confine yourself to 2d plot when you could visualize 3d plot with time as the z-component? But sticking with 2-d, choosing a single node as the center of gravity, you could have concentric circles to show the propagation of time and incremental node additions would appear within a time band. As you mention though, these only work well for smaller graphs as the edges will clutter the view (and look like a tangled spider web

A lot depends on the specific objective of what you’re trying to visualize and the insights you’re hoping to glean from such a visual.

Cheers.

Similar to this but replay/interaction with data is a must:

Sp3ar0

1. Line thickness = Log (t).

The thicker the line, the longer it has occurred.

2. Line colour. 8 hues for blocks of n hours to show last 8n hours. So if n=3 then last 24 hours.

I believe the above can be easily coded in without any change to top link graphs.

Cons: When there are > 20 sub-nodes linking to one node it will get messy. Otherwise if it is less than 10 it should be pretty discernible.

3. Line type. Tight and loose dashed lines to indicate how recent or how long event has passed. Faint line for events that occur more than 24 hours, eg. Again can be used with 8 or less colours to depict blocks of time it has passed.

Cons: When too many event sub-nodes occur more than 24 hours, line can be too faint to be discerned in a congested area.

Bottomline: Almost every implementation is bad if there are too many nodes/sub-nodes squeezed in a small area.