Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.misc > #7402

Re: [CM] Troll spotting algorithm

Cross-posted to 2 groups.

Show all headers | View raw

On 2015-04-14, Hils <hils@saynotospam.net> wrote:
>> And the ensuing battles will be known as the Troll Wars.
>
> It could be a way of developing AI systems, with troll bots and troll
> detectives playing cat and mouse among the human population, scoring
> points for evading detection or soliciting human replies, losing
> points for false accusations. They could be the Turing Troll Wars.

That would make this classic XKCD germane:
https://xkcd.com/810/

"Spammers are breaking traditional captchas with AI, so I've built a new
system.  It asks users to rate a slate of comments as 'constructive' or
'not constructive' ... "

I /do/ miss the days of really clever trolling: subtle remarks designed
to casually cause a conflagration.  These days so much of it is simple
ass-hattery.

I thought a bit about what this type of algorithm would look like: the
science ought to be something like Bayesian filtering, probabilities of
word order, and so on.  But it's hard to pick out sarcasm, nuances of
humor, and the like.  Post length is no easy indicator.  No single
vocabulary word gives it away.  IP address might help these days now
that so few are on dial-up and user addresses stay the same for longer.
On Usenet, user name often helps (ignore any post from a user named
"obamasucks@donkeyballs.org" or whatever on the grounds a fake ID with
an inflammatory address is likely to be out to stir the pot?).  You
could look for patterns of a user consistently replying after a post
about certain subjects (which would draw out the single-issue haters).
Beyond that, this isn't easy stuff and I'd think the false-positives
would be ridiclous. 

There are frequent, long discussions on soylentnews about how to tweak
the karma and scoring algorithms, but it always comes down to users
trying to adjust the math so they hear only from people they want to and
are barred from seeing posts from people they don't want to hear from.
Usenet, for all its shortcomings, does away with the issue by letting
anyone post anything and leaving it to the user with his killfile and
filters to sort out the damage.  Maybe the only way to win the game is
not to play it ...

Back to comp.misc | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

[CM] Troll spotting algorithm RS Wood  <rsw@therandymon.com> - 2015-04-14 14:52 +0000
  Re: [CM] Troll spotting algorithm Dan Espen <despen@verizon.net> - 2015-04-14 14:02 -0400
    Re: [CM] Troll spotting algorithm Hils <hils@saynotospam.net> - 2015-04-14 19:19 +0100
      Re: [CM] Troll spotting algorithm Dan Espen <despen@verizon.net> - 2015-04-14 14:28 -0400
        Re: [CM] Troll spotting algorithm Hils <hils@saynotospam.net> - 2015-04-14 22:11 +0100
          Re: [CM] Troll spotting algorithm RS Wood <rsw@therandymon.com> - 2015-04-15 11:26 +0000
            Re: [CM] Troll spotting algorithm Bud Frede <frede@mouse-potato.com> - 2015-04-23 12:42 -0400
  Re: [CM] Troll spotting algorithm Shadow <Sh@dow.br> - 2015-04-14 17:20 -0300
  Re: [CM] Troll spotting algorithm Andy Burns <usenet.feb2014@adslpipe.co.uk> - 2015-04-14 22:10 +0100
  Re: [CM] Troll spotting algorithm Bruce Horrocks <07.013@scorecrow.com> - 2015-04-15 11:02 +0100
    Re: [CM] Troll spotting algorithm Dan Espen <despen@verizon.net> - 2015-04-15 11:45 -0400
  Re: [CM] Troll spotting algorithm Sylvia Else <sylvia@not.at.this.address> - 2015-04-16 23:15 +1000
  Re: [CM] Troll spotting algorithm Bud Frede <frede@mouse-potato.com> - 2015-04-23 12:30 -0400

csiph-web