[Search-l] New idea - Based on Joel's one

[Marc .]

Joel
Your "scoring concept" seems to be a good way (and powerful concept).

I am thinking that is possible to combine this concept with a peer to
peer network.

Explaining:
If we make a algorithm that discover who the user "is" based on his
"scoring" (or any other behavior) this algorithm can determine an
address for the user in a peer to peer network based on his own
scoring.
In this way, users that look likes each other should be next to each
others, "naturally" clustering the users -assuming that the peer to
peer network accept only one address per user.

That should allow users to query other users and receive more relevant
answers to their queries, by classifying the answers received based on
how "near" the user who answer a query was from querier and how does
the link was scored by this "near user".
Note that this should allow, in principle, to receive the more
relevant answers faster too.

For example:

If I look for "jazz": I send a query "jazz" to my peers, they will
send me classified results (if they have one) and they will pass my
query "jazz" to an other peers.
Supposing that I receive only 2 answers to my query "jazz", but one
comes from a peer that I am directly connected to (1st degree), and
the other one comes from a peer of my peer (2d degree), I can say that
the first one it more relevant for me, because I am connected to users
that looks like me.

In other words:
I will probably receive, as result, first, and in a higher place, the
links concerning  "places to listen jazz in my city" because people
like me, when they think "jazz" they think "places to listen jazz" and
people like me lives in my city, so they view sites about "places to
listen jazz in the same city as mine" and probably have "good
classified links" in this subject for me -and they share it with me.
Note that, If I am an historian, I will probably receive first links
related to "history of jazz", its origins, influence etc.
In the other hand, if I am a student of music, I will probably receive
first links to "how to play jazz in saxophone" an so on...

Resuming the idea in 5 steps:
1- An algorithm try to "know who you are"
2- The peer to peer network puts you with "who you looks like"
3- You query a link to your peers
4- You classify the links by yourself helping the algorithm to "know
who you are".
5- You share your classification with your peers,  contributing to the
"Wikia Search"

Note that even the algorithms used to "discovers who you are" could
be, in theory, "auto modified" in the same way: if the you "click" (or
prefer) links proposed by users that uses algorithms "A" instead of
"B", that indicates that algorithm "A" should be better to you to
classify the links than algorithm "B" -who is running on your
computer- so algorithm "B" should be change to algorithm "A".

Ok,
I suppose that's enough for a draft of an idea! Note that I don't know
how to put this in practice ; ).

What does everybody think?

BR
Marc Rosenfeld