Normally, the value of a tag cloud is that readers choose the tags that are useful to them. Thus, the bigger ones are useful to more people.
These clouds are drvien by frequency of the candidates using words (I assume prepositions and definite articles were excluded). This approach does not allow anyone to choose tags based on utility, just frequency of use.
While it is interesting that some words were used frequently, most of the leading words offer minimal insight into the focus of the candidate.
What did jump out at me were coincidental sequences in the cloud, like Dodd's "Support terrorism today" and Gravel's "president respect (skip 1) soldiers."
Overall, this is curious with potential, but not yet analytically valuable.
Alan Rosenblatt
Executive Director, Internet Advocacy Center
AKA DrDigiPol (drdigipol.com)

print
email
delicious
digg
technorati

Visualizing Data
I love this and did it just a week or so ago to Bush's Iraq push back speech and Alberto Gonzales's testimony.
I tried it for the first time last year using Bush's speech commemorating the 5 year anniversary of 9-11.
I think the key is that "tag crowd" program allows you to insert hyperlinks for each tag. So if you were a really savvy campaign, you would click, let's say "terrorism" and it would take you to a page that said "my opponent says x,y,z about terrorism, but here is my position".
That would be a good policy, messaging use of a whiz bang visual technology.