Graphics Reference
In-Depth Information
they are in a sea of purple links. These are the most-watched television
channels, such as VQRI and VDRU, which are major national broadcast
networks (station identifiers have been replaced with these generic
names). Also, because they are very close to each other, the person is
most frequently clicking from one of these channels in the center to
another adjacent channel.
• Immediately above and to the left of the largest nodes is a cluster of
medium-sized purple nodes (for example, GAN, GANPE, GANZE, and
so on). This is a collection of specialty cable channels that have a high
degree of connection between them, forming a distinct cluster.
• Tiny nodes near the perimeter of the image surrounded by a lot of white
are stations with low viewership. They tend to have very few, very long
lines connecting them back to the rest of data—meaning that these are
channels that are not clicked through from very many other channels.
• Long tendrils of channels radiate out from the center. These tendrils
represent channels that are most commonly clicked through in a
sequence, and the start of the clicking sequence is more likely to start at
the end of the tendril closer to the center. The many purple links to and
from various nodes indicate many paths to and from these channels.
These tendrils tend to be channel packages (for example, a set of
basketball channels or a set of baseball channels).
To get a nice force-directed layout requires using one or more force-directed
layout algorithms. Some graph packages provide a number of different
algorithms with many settings. Some use the names of researchers; some
use generic names.
Some common layout algorithms may include the following:
•
Fruchterman Reingold
—The
Fruchterman Reingold layout
algorithm can be a good place to start. It lays out the graph in a fairly
compact circle, and separate components stay within this circle, as
opposed to flying away as found in some other layout algorithms. The
downside to Fruchterman Reingold is that all nodes tend to be the same
distance apart, and separate components may not be readily apparent.
Fruchterman Reingold is highly dependent on the layout before it is
run, so a better result may occur after first running one of the other
layouts.
