I got this from Steve GiBson (author of SpinRite)
and thought some of
you might be interested since you're Linux fans. I haven't tried the search
engine that he's talkng about so I have no idea if his claims are factual.
"Citation/referral frequency" may be the ultimate system of relevance
ranking for Mr. Gibson, and I'm glad that he has found the right system
for him to find the most POPULAR sites of what he wants.
But when you are looking for unusual/rare information, there might not be
ANY other sites that point to what you want, and popularity becomes an
inappropriate ranking. For example, with his system,
Amazon.com would
always come out ahead of
Abebooks.com
If you want exactly the same as what everybody else is looking for, fine.
But if your needs do not match the MODE, optimization for RECALL is most
important, and reduction for PRECISION can be done ONLY with sufficient
RECALL. Therefore, a probabilistic or even vector-space ranking would
work better. But for MY needs, carefully chosen search terms in DOGPILE
works best.
--
Fred Cisin cisin(a)xenosoft.com
XenoSoft
http://www.xenosoft.com
2210 Sixth St. (510) 644-9366
Berkeley, CA 94710-2219