Algorithm spots internet trolls after 5 posts, offers 80% accuracy

[ik9000]

Rubbish!

[ik9000]

Yup, some sort of override going on there...

[ik9000]

You realise those posts would have got me autobanned? That is why I am cautious about forum owners hiding behind faceless algorithms. Yahoo have already trimmed tech support to what seems like 1 man in a caravan on Wednesday lunchtimes between 1-3pm and only on the wane of the lunar cycle. Algorithms will get wielded, and there will be no-one to unban the legitimate users!

[Xlucine]

If the mod is happy to give free reign to an algorithm with a 20% false positive rate, are you sure you'll enjoy discussion there? If they're that trigger happy, then it'll only be a matter of time before every poster has annoyed them somehow and they end up with 0 users.

[MaddAussie]

rubbish!

[miniyazz]

Accuracy is such a meaningless term. Need specificity and sensitivity at least to be able to make any informed decisions.

I can block troll posts with 100% accuracy automatically, oh yes it does involve blocking every post made, why do you ask?

[Otherhand]

I have my doubts about this. I think it can possibly spot people with compulsive aggressive tendencies, but that's not the same thing as a troll. A good troll will seem like one of the better community members at first and suddenly turn sour some posts later, or better still they'll dance very close to the banhammer while confusing everyone about what they're doing and avoid a ban for as long as possible.

When I first saw the headline, I thought it was going to be something cleverer than this: maybe a pattern in posting frequency, message lengths etc.

[Saracen]

Firstly, I'd be interested to hear what the Hexus lords think of this - if this was available to Hexus would they use this automated troll hammer?

.....

As an alert, maybe .... if it worked well enough.

But 80% isn't accurate enough, or even close, IMHO.

I'd want to be satified that, first, it worked effectively, and that it didn't create more problems with false positives than in solved with accurate positives. But second, that it's sense of what is, and isn't, actually trolling was sufficiently refined. And I struggle to see any automated filter doing that well enough.

I'd consider it, as an alert system, but am VERY sceptical.

[royo]

But 80% isn't accurate enough, or even close, IMHO.

Indeed. Out of 100 users you're going to ban 20 for no reason with this algo.

[peterb]

Indeed. Out of 100 users you're going to ban 20 for no reason with this algo.

It doesn't need to be an automatic ban, it could keep a post/user in limbo until it had been reviewed - an aid or, as Saracen says, an alert.

[Agrippa]

I think it can possibly spot people with compulsive aggressive tendencies, but that's not the same thing as a troll.

While I agree they may be perceived as different if one can be bothered to dig deep enough, for all practical purposes they are the same thing. And both are better out than in.

A good troll will seem like one of the better community members at first and suddenly turn sour some posts later, or better still they'll dance very close to the banhammer while confusing everyone about what they're doing and avoid a ban for as long as possible.

True enough, but trolls of that "quality" are very few and far between. I've come across maybe five or six in my time on the internet, which began when it first became commercially available. Its also quite entertaining to watch a master at work and given that and how endangered they are as a species I'd be inclined to let them roam free.

Sadly 99.9?% of trolls appear to simply be dimwits with bad or no manners and poor impulse control.

[abaxas]

Also remember that computers and autism are usually good friends. Hence there is a fine line between calling out a troll and being a right **** yourself.

[kalniel]

Also remember that computers and autism are usually good friends. Hence there is a fine line between calling out a troll and being a right **** yourself.

And why some people go to the effort of creating an algorithm to call them out instead of doing it themselves

[crossy]

Yahoo have already trimmed tech support to what seems like 1 man in a caravan on Wednesday lunchtimes between 1-3pm and only on the wane of the lunar cycle. Algorithms will get wielded, and there will be no-one to unban the legitimate users!

Yahoo still exists? I was going to go all Zen and ask "if a Yahoo admin bans you, would you even notice?"

Actually, when we say "troll" am I right in thinking that we're speaking about those barely literate keyboard monkeys who take an unreasoned position and then try to "justify" with invective?

Reason I ask is that I'm currently having a "debate" with someone on the merits (or according to him - total lack) of the output of the group AC/DC. I've described him as a "troll" but maybe I'm being unfair and the guy is just a moron. (If you're curious, my opposite is taking the stance that AC/DC's entire song catalog only uses 3 chords - despite me and others pointing out that this is patently untrue. He's now moved onto "... and it all sounds the same").

Sorry for the digression, but I'm still hacked off that the Hexus website seems to have reverted to not sending me a daily post summary (again). Good start to the day....

[DanceswithUnix]

Yeah, depends how it is used. If it points mods to users that need a careful eye then that is probably a time saver. If it is the equivalent of speed cameras, where highway police are replaced with a device that only monitors one specific aspect of driving making the world a worse and more dangerous place then it would be a good way to drive people away from a board.