Re: [GNUnet-developers] Re: amortizable hashcash paper

[Igor Wronsky]

On Fri, 9 May 2003, Tom Barnes-Lawrence wrote:

> -Finally: The proposal emphasises that the votes be used to *sort* the
> search results. On the one hand, this is quite a good thing that it's
> not talking about censorship as such, *but* think about it: Everybody
> searches for foo. First person checks the first 10 results, finds them
> OK, and votes for them. Next person is more likely to get those
> results sooner, and vote for them. And the next person. The voting
> system could quite easily polarise the popularity of search results
> such that certain results would usually only appear nearer the far
> end of the lists and stay there.

That seems a serious problem. And it reminds me that we already
have a problem of keyword saturation. Its basically 'first
come first served' type of problem. If someone inserts something
under a typical keyword, and people search for that key, the
results propagate across the network. If later on someone else
inserts something, the probability that it reaches the querying
node is smaller because globally the older content already
dominates in quantity. As this process goes on, it becomes
increasingly difficult to put anything under any reasonable
keyword, because the keywords are saturated with old content.

One option would be tag each possible answer with a local
counter (not ever sent over the network) that was increased when
the particular answer is sent across the network as an answer
to a query. Then the randomized selection could be inversely
biased by this value: prefer to send such replies that have been
sent less often. It would give fresh content a greater chance
at propagation. A further point: keyword hits in the db are
scored/prioritized by keyword. They don't make distinction
'what' it is that the keyword refers to, at all, so if someone
has spammed a keyword and those spam-hits have widely propagated,
it doesn't say anything about people wanting just them, only
that they're interested in the particular search key.

Of course this doesn't prevent that eventually the reasonable
keywords will still be dominated by obsolete i.e. "past" content
that might be rendered useless by technical advances (for
example low resolution files).

This raises a question if the search mechanism as it is has
asymptotically much value. ;)


Igor