On Mon, 28 Aug 2006 09:33:48 +0200, Big Bill <kruse@cityscape.co.uk> wrote:

> "In an implementation consistent with the present invention, the set
> of favored and non-favored sources may be automatically determined. To
> accomplish this, exemplary queries in the query theme may be
> classified into a set of topics (e.g., an online topic hierarchy, such
> as Yahoo!, Open Directory, or Google) using the approach for
> classification described above. Web hosts that appear in the URLs
> associated with the best matching topics to the query theme may be
> taken to be favored sources. For example, if the query theme is "sites
> that in finding accommodation," then web hosts listed under the
> Open Directory category "http://dmoz.org/Recreation/Travel/Lodging"
> can be taken as favored sources."
>
> Seems like Google are admitting the limitations of their algo (no big
> surprise to the rest of us) and trying to compensate by bringing in
> human supervision. Which, to a degree, they've always done by relying
> so heavily on DMOZ to function as an ancillary human judging system.
> Problem is, now that they seem to be wanting to expand that aspect,
> who they gonna call? DMOZ is pretty much a spent force and if they
> hire people off the street then how are they going to train them? Who
> will judge the judges?

That's nothing else but Trust Rank, old news in a way. Idea is that it is
enough to select small group of trusted sources, thus they don't need huge
team of judges. Nodes/sources (like DMOZ) can be selected by algo, then
you just need one person to browse them now and then to check if they can
be (still) trusted. I don't remember exact details (and they even
evaluated how many sources you need), but I have seen Trust Rank paper
available somewhere on the net.

Best,
Borek
--
http://www.chembuddy.com
http://www.ph-meter.info/pH-meter
http://www.terapia-kregoslupa.waw.pl