tedster - 6:06 am on Feb 8, 2007 (gmt 0)

Yes, I'd say the new Spam detection patent [appft1.uspto.gov] is a good candidate. It certainly does re-sort the results and it also has the potential to generate drastic changes for just some search phrases and not others.

I pulled out a few key sections.

(Note: "good phrase" here means topically predictive) [0218] ... If there are a minimum number of good phrases which have an excessive number of related phrases present in the document, then the document is deemed to a spam document. [0223] If the document is included in the SPAM_TABLE, then the document's relevance score is down weighted by predetermined factor. For example, the relevance score can be divided by factor (e.g., 5). Alternatively, the document can simply be removed from the result set entirely. [0224] The search result set is then resorted by relevance score and provided back to the client.

In addition, the inbound anchor text scoring described in the patent could boost the document out of the danger zone if even a single new IBL shows up.

You can get your page included in the spam table by having too many occurances of semantically related phrases. This certainly suggests some modifications for both body copy and anchor text that might be worth an experiment.

The threshold for "too many" can also be fine-tuned in this algorithm, which would create the "popping in and out of the penalty" quality that some people report.

[edited by: tedster at 8:18 am (utc) on Feb. 8, 2007]