> Well, I'd have to rate the books I read for it to
> have something to match with other folks' ratings,
> right? Or am I missing something?
yes, right. but that's not "work", it's _fun_. :+)
i'm totally serious. i've written the program that
collects such ratings, and it's lots of fun to run.
being reminded of books that you've read, and
tasked with giving each one an overall reaction,
is a pleasant thing (even if it's not all that easy).
the rating scale runs the full width of your screen,
from "1.00" on the one side to "9.99" on the other;
you just click at the appropriate spot in the middle.
(a rating-scale this fine-grained helps to avoid ties.)
your ratings are then ordered, so you see the list of
books (or songs or films or whatever) and can drag
each of them up or down in the list if you want to...
it's so entertaining, in fact, that i'm gonna release this
rating-program just for the fun-value, as soon as i can
find the time to put together an impressive list of books.
so if anyone has some pointers to lists of books, tell me!
i want my list to be composed of a wide variety of books.
i also want it to be an extensive list, with _lots_ of books,
because one part of it will be to formulate a ratings profile
after having obtained judgments from a subset of the list.
then, as the task continues with the rest of the books,
using your preliminary ratings profile, when the program
puts up a book-name to get your rating of it, it will also
display its _prediction_ as to what your rating will be...
when you realize that these predictions are _accurate_
-- and that they get even _more_ accurate as you give
more and more ratings to the program -- you will be
convinced and amazed that this shit really _does_ work.
and given this excellent performance in predicting your
ratings on the books that you've already read, you will
have great confidence in its predictions for the books
that you haven't yet read, and will take them seriously.
i forgot up above to say (clearly) that when you get your list
of recommendations from the system, including its prediction
of the rating that you'll assign to each book after you read it,
those predictions will be uniquely tailored for you personally.
so, for the very same book, it might predict that i'll rate it as
an 8.42 (since that's how people with ratings similar to mine
have rated that book), while it predicts that _you_ will rate it
as a 4.21 (because that's how people with ratings like yours
have rated that particular book). so it will _recommend_ that
i will _like_ the book, and predict that you will _not_ enjoy it.
this was probably obvious to you, but i wanted to stress it,
because when we look at amazon's recommendation system,
we see it doesn't custom-tailor suggestions to each person.
it can't, because it doesn't know anything about _your_ taste.
when it says "people who bought book x also bought book y",
it's talking about those other people and not about _you_, so
it presents that exact same little factoid to every one of us...
and although we are tempted to infer that this means that
book x and book y are "similar", the _basis_ of the similarity
might be totally bogus. maybe the reason people are buying
both of these books is because they're both being advertised,
heavily, and the buyers are sheep doing what they've been told.
and maybe every person who loves book x also hates book y,
while everyone who loves book y hates book x, we don't know.
and amazon doesn't care, because they got your money for both.
if any particular variation of a "collaborative-filtering" system
gives the same output to _everyone_, then it's _not_ gonna do
a good job of pulling the needles out of the haystack, because
there is too much variation in human taste for one size to fit all.