norton_j_radstock - 8:14 am on Aug 26, 2006 (gmt 0)

This is a really significant move on the part of Google -in essence it is similar to teaching a computer to recognise your handwriting -you don't need to instruct it on every piece of information, but rather just give it a big enough sample.

So how might this work? Let's say a team of human editors are scoring sites according to the Google manual and so called 'thin affiliate' or scraper sites are being given low scores. The computer then works out what the common factors are for those low scoring sites and makes that part of the algorithm.

Thus in simple terms it might calculate, for example, that scraper sites are commonly composed in the form of a directory of links followed by 20-30 words of text and a line break, all repeated several times, and with a high proportion of those links pointing at similar affiliate URLs. Other sites sharing similar characteristics would then be marked down in terms of ranking of search results.

Inevitably there will be good sites that fall victim to such systems, the ultimate aim of such a system is continual improvement aided by the ongoing human input. The important thing is that the humans are unbiased, so it is far more likely that Google will continue to employ trained expert assessors, rather than leave it up to the web at large.