Re: Re: My Gnutella spam reduction idea revised

Courtney,

What would happen if the user searched for "flowers"?  According to
your scheme, the program would shorten it to "flower", and then all
clients that returned a result containing only "flower" would be
considered spam.  Should we condemn ourselves to a life without a
flower?

This is not to say there isn't merit behind the idea, but merely that
implementing without banning half the network as spammers is a lot
more tricky than just dropping a letter.

Sam

On 4/25/07, just_courtney_girl <[email protected]> wrote:
> Well, only the spambots would answer.  They
> would look at the query, create a file (or
> the appearance of a file that can be created
> if someone bites), and return a result,
> regardless of the string.  But that is not a
> help.  Even if we knew they were spammers
> at this point, what could we do if requested
> cargo didn't also exist?  Even if the data
> returned is from the spammers, would it be
> reliable enough to exclude for so many
> searches?  Plus, why send good searches
> after bad ones?
>
> The only work-around I'd see would be another
> search excluding the clients that answered
> the first time, and that is the very situation
> we should try to avoid.  If we do a second
> search, that is eating up bandwidth and there
> is no guarantee that the clients marked for
> exclusion are the spambots.  They could
> simply be clients that relayed the hits or
> machines of unfortunate souls who were
> tricked into downloading the junk.
>
> Just some thoughts.
>
>                          Courtney
>
> --- In [email protected], "Zlatin Balevsky" <zlatinb@...> wrote:
> >
> > Very interesting thought, Michael.  The only issue is
> > backwards-compatibility.  Current clients will not respond to a multi-word
> > query unless all words match.  If you add a new made up word, they will not
> > answer at all.  Do you have any ideas how to work around that?
> >
> >
> > On 4/23/07, Michael Rogers <m.rogers@...> wrote:
> > > Hi Courtney,
> > >
> > > Sounds like a good idea in principle, but I believe some Gnutella
> > > clients match partial words to provide more intelligent reponses to
> > > things like plurals, common suffixes and apostrophes. Instead of
> > > chopping up one of the query terms, how about adding an extra query term
> > > that's unlikely to occur in any real hits (eg a made-up word)? It's easy
> > > to use a letter/digram/trigram frequency table to generate
> > > realistic-looking words (too realistic, actually - you'd probably need
> > > to filter them through a dictionary to remove any real words that were
> > > generated by accident).
> > >
> > > Just a thought,
> > > Michael
>
>
>
>
> Yahoo! Groups Links
>
>
>
>