Re: [circle] Voting on p2p

[thomasV1]

> Jiri:
> 
> > One issue is security: how do you prevent one person
> > from artificially increasing the ranking of their files?
> 
> Indeed, that seems to be the hardest problem right now...
> 
> If we can't think of anything better, one possibility would be to simply 
> ignore this problem for now, and see what happens. Circle is vulnerable to
> intelligent attackers anyway...

as long as there is no financial reward associated
with the ranking, I guess nobody would care trying 
to cheat, indeed

> Umm, I was thinking of a more concrete algorithm for aggregating votes, 
> really... the mechanics of getting votes from A to B are easy. (My initial
> thought was to simply have them in a file with a standard name, shared
> like 
> any other file, which would have the advantage that it could be
> retro-fitted 
> to *any* file-sharing program, not just circle. The other methods proposed
> would work just as well, though.)

that is simpler indeed.

> 
> The part of the aggregation method that I have least thought through is
> the 
> bit where I say "the people who tend to agree with you on the rankings".
> How 
> is this measured? Given two people's votes, how do you determine whether 
> their tastes are similar?

most of the time two people will have given their opinion 
on different music tunes, so you cannot directly compare 
their vote vectors, because there will be little or no overlap.

A method is to look for pairs of people whose votes overlap. 
Create a big graph, where two people i and j are linked iff 
their votes overlap. Assign this edge a value W_ij (weight) 
that is the number of overlapping votes.
Assign this edge another value that is the distance D_ij
between their votes, restricted to the overlapping values.

now if you consider 2 nodes i and j that are not linked,
you will have to find a path between them. more precisely, 
you will need to consider all the possible paths between them,
and integrate over them while favouring shorter paths.

\sum_{path in possible paths between i and j} 
\sum_{(k,l) in nodes of path} f(W_kl,D_kl)

where f is a big nasty complicated function that must be of
bayesian inspiration, that takes into account the fact that 
you want to prefer shorter paths, and that tells a lot 
about your tastes, childhood and and political views.

hum, the above algorithm is of course patent pending.
please contact me for licencing and legal issues :-)


-- 
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgebühr: http://www.gmx.net/dsl