tedster - 3:31 am on Sep 2, 2010 (gmt 0)
We have several current threads here mentioning that the old "tried and true" on-page SEO methods for Google don't seem to be working the way they used to - for example:
Updating Question - no ranking movement despite on-page changes [webmasterworld.com]
Exact Long Tail Phrases in the Title are Not Ranking Well [webmasterworld.com]
There's a scholarly document available from Penn State that describes a potential type of analysis that may be at work - Hidden Topic Markov Models, or Latent Dirichlet Analysis. Yes, the math is a bit intense but it's worth a scan for anyone with an appetite for understanding the kind of tech that may be folded into the Google algorithm these days.
Here's a URL for the PDF file, a paper from Amit Gruber, Michal Rosen-Zvi and Yair Weiss:
Yes, some pages are ranking well today with only related vocabulary rather than directly plugging in the search term keywords. And other pages rank better when the related keywords are added. It's a mysterious effect for anyone who is used to a more literal type of search result, whether this is the exact math being used or not.