Douwe Osinga's Blog: More on LandGeist

Tuesday, January 13, 2004

More on LandGeist

A week ago, I wrote about Landgeist . Landgeist projects for specified keywords the correlation between the country name and the keyword on a world map, i.e. coloring Iraq very red for War and Colombia for Coffee.

I choose Alltheweb because Google doesn't seem very good at counting results. 8% of all pages with "united states" contain "cheese", according to Google. Other search engines put it around 1%, which sounded more reliable. The maps looked good, but a little funny:
Landgeist - Alltheweb

For example, Iraq scores very well on War and Islam, which seems fair enough, but also on Sex, which is slightly weird. The best explanation seems to be that Iraq occurs in discussions with lots of other words. Using Altavista with the NEAR keyword gave much better results, because it only counts pages where the name of the country and the term occur next to each other.

So here it is, a new version of Landgeist, this time using Altavista: