Minority Reporting your comments section
May 8, 2015 1:23 AM   Subscribe

In general, we can conclude that trolls of all kinds post too much, they obsess about relatively few topics, they are often off topic, and their prose is unreadable as measured by an automated index of readability. Readability was one of the strongest predictors they found. They also generate lots of replies and monopolise attention.
Alex Harrowell summarises the results of a study into automated predictions of commenters most likely to turn trolls (PDF link to original paper).
posted by MartinWisse (59 comments total) 22 users marked this as a favorite
 
So, the moral compass of cortex has finally been reduced to an algorithm?
posted by ouke at 1:26 AM on May 8, 2015 [7 favorites]


This might explain my general lack of success when trolling football forums. I'm simply too coherent.
posted by brokkr at 1:41 AM on May 8, 2015 [3 favorites]


Metafilter: they post too much, they obsess about relatively few topics, they are often off topic, and their prose is unreadable
posted by kagredon at 2:21 AM on May 8, 2015 [57 favorites]


The early intervention seems to cause increased escalation was fascinating to me - I'd be interested to see how this plays into the motivation for trolling and what reinforces trolling.
posted by Deoridhe at 2:27 AM on May 8, 2015 [3 favorites]


I'd be interested to see how this plays into the motivation for trolling and what reinforces trolling.

(Too much free time + too little creativity) x too much easy attention = troll
posted by Alexandra Kitty at 3:37 AM on May 8, 2015 [5 favorites]


trolls of all kinds post too much, they obsess about relatively few topics

Plus ça change, etc. Microsoft did some work on Usenet behaviour ages ago, and provided a nice API with some aggregated poster metrics that I used to build a usenet troll checker. The algorithm was really simple:

"The R number is the ratio between replies and posts; if most of your posts are replies, chances are you’re either a troll or a helper. The T number is the ratio between threads and overall posts; if you spend most of your time posting to the same thread, you might be a troll."

It was often quite accurate, at least for the groups I followed.
posted by effbot at 3:50 AM on May 8, 2015 [4 favorites]


Huh: "trolls with more cultural capital adopt strategies of disruption that allow them to persist longer and do more damage."

That would explain the failure modes that Mefites still sometimes display.
posted by anotherpanacea at 4:01 AM on May 8, 2015 [5 favorites]


But effbot, how can that be? I mean, you've made over 1100 replies here, but no posts. If your algorithm works, what does that say?
posted by Kirth Gerson at 4:03 AM on May 8, 2015 [3 favorites]


I think it's more complicated. The perception of who is a troll is subjective. Most Mefites would be considered trolls if they tried to voice their opinions on a site like lucianne.com.
posted by Pararrayos at 4:14 AM on May 8, 2015 [1 favorite]


The perception of who is a troll is subjective.

I'd say that more like, "Trolling is context-dependent." If you want to troll Metafilter, you might use pick one type of comment on, say, gender and sexuality, but you would have to use very different comments to troll a different community. The trollish behavior remains constant, but the form it takes has to vary in order to get the desired reaction.
posted by Dip Flash at 4:18 AM on May 8, 2015 [8 favorites]


If your algorithm works, what does that say?

Yeah, the R number indicates that I'm either a troll or a helper. Luckily, there's one more condition.

(which is why I usually limit myself to 3 comments per post I'm commenting on, to be on the safe side :-D)
posted by effbot at 4:25 AM on May 8, 2015 [1 favorite]


chances are you’re either a troll or a helper.

A couple professional mailing lists I'm on include trollish helpers in their membership — folks who honestly want to help and who respond to many requests for help, but who cannot avoid beating their dead hobby-horses at the same time and who do not take criticism well.

One of them in particular is quite helpful answering questions related to the way things are currently done in the mailing list's topic — but he also disrupts any substantive discussion of the upcoming new approaches.
posted by metaquarry at 4:36 AM on May 8, 2015 [1 favorite]


Thinking about this some more, I can't help wondering if all the algorithm does is determine the criteria moderators are using to ban people. So it's not necessarily the case that the people being banned are trolls, just that in the average community there's a list of behaviors that throw up red flags and get you booted.

This list of behaviors is pretty generic (thus the high false positive rate) and it doesn't speak to intention: it's a prediction about who has become so annoying the moderators are likely to intervene. It's also notable that "trolling" is only based on the last 8 to 10 posts: a person's long term history with a site or the topic is ignored, and indeed it's ignored both by the moderator doing the banning and by the algorithm, equally. The kind of communities that ban users after a bad night (instead of timeouts) are using something of a hair trigger and targeting behaviors not people.

In that sense, a community with a very different demographic, like Metafilter, probably couldn't run this automated troll identifier usefully: for us, the average troll is someone who is single-minded and fights over much longer than 8 to 10 posts. And intentions become more evident in that time, so we can afford to develop a sense of someone's sincerity. Or else it's a marketing person, and so not a troll but a spammer.

In any case, I don't think there's much of a connection between the way you troll football fans and the way you troll Mefites. It seems like we're more easily trolled by infrequent provocative drive-by commenters, for instance.
posted by anotherpanacea at 4:44 AM on May 8, 2015 [2 favorites]


I miss the days when the word "trolling" was useful because it referred to a specific kind of disruptive activity (and a way of catching fish), instead of just to any kind of online behaviour that the person using the word regards as undesirable. For one thing this seems to mean that people don't know actual trolling when they see it and tend to respond by giving the troll all the food it wants, something we see on the odd occasion that Metafilter gets trolled.
posted by A Thousand Baited Hooks at 4:46 AM on May 8, 2015 [9 favorites]


posted by A Thousand Baited Hooks

Your presence in this thread alone is eponysterical.

They trained their algorithm at Breitbart and CNN: has anyone ever actually visited those comment sections? I'm not brave enough. It seems possible that their definition of a troll is exactly what we'd think of as informed discussion, especially Breitbart which is basically an entire site devoted to conservative political spin and provocative reporting.
posted by anotherpanacea at 4:56 AM on May 8, 2015 [3 favorites]


I think it's more complicated. The perception of who is a troll is subjective.

more specifically, the perception of being a troll depends upon both the median opinion about a topic and the strength of the peak around the median, if you can imagine distributing all opinions on a given topic on a binary scale between two opposites.

on a forum where opinions cluster very strongly around a median viewpoint i.e. everyone thinks pretty much the same, a viewpoint even slightly away from that peak is heedlessly provocative. perversely, the more you expunge opinions away from the median, the stronger that median becomes and the more sensitive the community becomes to "extreme" opinions.
posted by ennui.bz at 5:03 AM on May 8, 2015 [5 favorites]


Yeah, "troll" doesn't mean anything any more.

Also, I predict that the usual Metafilter Foghorns, Busted Urinals and Sea Lions will instinctively steer clear of this thread, whether they realize it or not.
posted by sidereal at 5:15 AM on May 8, 2015


Anyone heard about a cheap way to score viagra?

If I troll this post, am I on topic?
posted by Nanukthedog at 5:20 AM on May 8, 2015 [6 favorites]


I find the real ranters can't spell or punctuate either.
posted by GallonOfAlan at 5:20 AM on May 8, 2015


Okay, I need to know what Foghorns and Busted Urinals are.

Unless you're trolling.
posted by escape from the potato planet at 5:28 AM on May 8, 2015 [2 favorites]


sidereal, though I may regret asking, what is a Busted Urinal?
posted by wenestvedt at 5:29 AM on May 8, 2015


Twenty bucks, same as in town.
posted by griphus at 5:33 AM on May 8, 2015 [12 favorites]


sidereal, though I may regret asking, what is a Busted Urinal?

fwiw, Google says "Did you mean: Best Urinals" which I guess means I hit their North Korean servers.
posted by effbot at 5:33 AM on May 8, 2015 [2 favorites]


People don't notice the best trolls are even trolls. I think with a focus on things like low quality prose or machine gun posting they are only going to get the obvious ones.
posted by Drinky Die at 5:39 AM on May 8, 2015


According to an archived post on comp.lang.java.programmer

Busted Urinal Award

The reward for the Lamest of the Lame on Usenet. One who clogs and stinks up
the joint like a busted urinal, you could say.


Foghorn is a reference to Foghorn Leghorn, I believe.
posted by metaphorever at 5:41 AM on May 8, 2015 [2 favorites]


They trained their algorithm at Breitbart and CNN: has anyone ever actually visited those comment sections? I'm not brave enough. It seems possible that their definition of a troll is exactly what we'd think of as informed discussion, especially Breitbart which is basically an entire site devoted to conservative political spin and provocative reporting.

I think given that the algorithm worked as well on the various sites shows that "troll" behavior has similar characteristics, regardless of the focus of the site. Still, I'd agree that the research's criteria for what makes a troll a troll could be subjective. (Which probably explains the 20 percent fail rate.)

As for the slowly progressing troll, it's probably a principled minority opinion holder -- the Christian conservative on the progressive site, the environmentalist on Breitbart. They usually steadily devolve as the group sets on them and tries to drive them out. I suspect these trolls wouldn't see themselves as trolls. And it does make you wonder who's to blame for trolls.

Which is what I'm interested in: why troll? It's more than just attention-seeking, I'm guessing. Do trolls actually serve a purpose? They do constantly challenge group dynamics. What's the effect of trolling? I feel like the trolls have essentially taken over the Internet -- and it's a much worse place for it -- but what does that mean?

Whatever. I'm just rambling now.
posted by touchstone033 at 5:44 AM on May 8, 2015


So, the moral compass of cortex has finally been reduced to an algorithm?

And more importantly, when can we deploy the Project: Insight Helicarriers?
posted by radwolf76 at 6:15 AM on May 8, 2015


I use Busted Urinal in the ancient Usenet sense, one who simply runs on and on and on, not loudly (like the Foghorn), not personally (like the Sea Lion), just blandly relentless, unaffected by (and seemingly unaware of) anything you say or do to them, including metaphorically peeing on them.
posted by sidereal at 6:19 AM on May 8, 2015 [5 favorites]


Busted urinal is sexist. I prefer "malfunctioning human waste fixture".
posted by dr_dank at 6:23 AM on May 8, 2015 [1 favorite]


I think trolling is a kind of bullying you feel justified in. Like when I'm mean to scammers on dating sites, and actively engage them and point out flaws in their argument/conversation from my point of view. There's a couple of key things there, the schadenfreude that comes from stirring someone else up to the point that they are clearly angry or lose control (yes, the misfortune but is caused by me) and the feeling justified, likely not because of the conversation topic, but because of who the trollee (trolley? Trolled?) presents as. If I were a men's rights activist (whatever colour pill that is), I might feel comfortable in disrupting a feminist conversation about rape, even if I don't think rape is good, because I don't like kind of people who are engaged in the conversation. Of course, it usually comes about that people with a certain overarching philosophy will disagree with the people who oppose it, but that doesn't mean they will troll each other. It requires a belief that the trolled one is undeserving of normative social niceties, that they are wrong or bad by the troll's own viewpoint. And the troll enjoys punishing people for that viewpoint.

This study doesn't identify what makes people troll, but it does show that when trolls are rewarded in their disruptive activities, by posters who are frustrated / moderators calming, that it increases their engagement. My own experience in trolling (mean-spiritedly too) OKC spammers supports this. When they get angry, I feel like a successful 14 year old who has increased status by excluding a girl out of a group.

Early shadow banning seems like an effective option in a public group. It allows the troll to keep posting but doesn't reward them with disturbed responses. Of course, this in itself seems unethical, to not inform a poster that they have lost the right to share their opinion, but it can prevent them escalating with multiple user names, as a result of the attention reward. I think Metafilter bypasses this to a large extent with the five dollar sign up. If you are banned and you return (through obscuring your IP), your behaviour will cost you another fiver and another fiver and so on.

The only problem I can see with this model is that it excludes the very poorest who have access to the Internet, but it maintains honesty in moderation.
posted by b33j at 6:35 AM on May 8, 2015 [3 favorites]


I think trolling is a kind of bullying you feel justified in.

That's what I came to believe eventually. I used to engage in some light trolling of some conservative sites, mainly aiming to satirize their views and promote thought more than anger, but even so I eventually just saw it as a form of bullying. Maybe you can make a case that some people deserve bullying, but I've always found that bullying is toxic to the mind of the bully.

Enjoy schadenfreude when it pops up naturally, if you try and create it it's just sadism. Taking pleasure in anger or hatred directed at others eventually gets a lot of people to start looking for more and more things to get angry at. Recreational outrage is a dangerous drug.
posted by Drinky Die at 6:49 AM on May 8, 2015 [9 favorites]


Isn't the key characteristic of trolling insincerity? That is, a troll is someone arguing or promoting opinions that they don't actually hold, or don't hold to the extreme degree that they present, specifically to anger people, "stir the pot", etc.

Someone sincerely arguing a context-outlandish position isn't a troll.
posted by Sangermaine at 6:52 AM on May 8, 2015 [1 favorite]


The definition has really shifted to any sort of disruptive or offensive behavior as far as I can tell, but yeah that's closer to the original definition.
posted by Drinky Die at 6:54 AM on May 8, 2015


It seems like the benefit of any automated troll-score thing like this is that your community doesn't waste endless time trying to divine the intentions of the disruptive person, and can instead moderate based on behavior patterns and data. For all the time we spend on "Is this person insincerely trolling us, or are they presenting genuinely held beliefs in a disruptive way?" discussions we have here, the answer is unknowable.
posted by almostmanda at 7:13 AM on May 8, 2015


Isn't it just obvious that a troll in the original sense of an insincere pot-stirrer is going to avoid algorithmic detection?

If there was something specific that looked like, it wouldn't work. In many ways the "concern troll" is axiomatic: someone who pretends to care even more and is just raising a potential problem for consideration. Sincerity is pretty easy to fake (that's what marketing is) and on the internet doubly so.

But since sincerity is easy to fake, we turn to behavioral evaluations. Yet that's where the high false positive rate becomes so troublesome: you end up misfiring too often, and the technique loses legitimacy. Depending on the base rate of trolls, a high false positive rate can lead to a situation where detecting real trolls is less common that false positive indicators unfairly labeling non-trolls. (Sometimes called the false positive paradox.)

In debates here there's been a lot of emphasis on not labeling other users "trolls" just because in practice the term just means anyone we find annoying. If anything produces more false positives than this automated prediction method, it's our own sense of outrage.
posted by anotherpanacea at 7:20 AM on May 8, 2015 [1 favorite]


Isn't the key characteristic of trolling insincerity? That is, a troll is someone arguing or promoting opinions that they don't actually hold, or don't hold to the extreme degree that they present, specifically to anger people, "stir the pot", etc.

Yeah, the meaning seems to have moved on, which I regret.
posted by Wolof at 7:26 AM on May 8, 2015


I also think there's a clarity of purpose implied in the word "trolling" that isn't usually there. I'm sure a lot of what we call trolling is initially sincerely stated but turns insincere when there's a severe reaction, or initially stated overbroadly and bluntly and refined in followup comments to sound more reasonable. Like Schrodinger's douchebag.
posted by almostmanda at 7:27 AM on May 8, 2015 [1 favorite]


The definition of troll is a huge begged question. I'm not finding any clear, meaningful definition of it at all. The study is about the sort of behaviors people exhibit prior to being banned. You could probably cast that as studying the behaviors of people being bullied and silenced or something if you were inclined.

Is any person who disagrees with a majority opinion a 'troll'?

I'm OK with the definition of 'troll' changing. (Not really, but I don't have a valid argument against it). But tell me exactly what it means. Be precise. I'm looking and looking and all I'm finding is circular arguments and incredibly vague assumptions.
posted by ernielundquist at 7:39 AM on May 8, 2015 [3 favorites]


Look, there's one now
posted by sidereal at 7:45 AM on May 8, 2015 [2 favorites]


Perfect!

(that was deliberate, right? Please tell me it was)
posted by A Thousand Baited Hooks at 7:52 AM on May 8, 2015 [2 favorites]


why troll? It's more than just attention-seeking, I'm guessing.

And considering how pointless so much of "writing on the internet" is, why put in the effort? Boredom? Sociopathy? A twisted sense of humor? I don't know. Lately, tho, I've been thinking about the libidinal/emotional relationship that nearly all of us (those of us in the Anglo and industrialized West, certainly, and increasingly globally) have with "the internet." We tell it our opinions and our secrets; we make it a mirror of ourselves and our families; we use it to gratify our demands and our fantasies. Our relationship to our screens and to what's "on" them is nearly as invested with our deep desires and sense of self (or any lack thereof) as any of our relationships with our parents, partners, or therapists. Maybe more so. "The internet" is the great collective dream/nightmare of the 21st century. If it seems insane at times, that's probably because it is.

(Cf. Goya, El sueño de la razón produce monstruos and Morbius, Edward, "My poor Krell.")
posted by octobersurprise at 7:59 AM on May 8, 2015


The trollish behavior remains constant, but the form it takes has to vary in order to get the desired reaction.

So, you're saying Trolls are Gozer the Gozerian.
posted by Pogo_Fuzzybutt at 8:01 AM on May 8, 2015 [1 favorite]


[insert off-topic Marx quote here]
posted by You Can't Tip a Buick at 8:02 AM on May 8, 2015


whoops sorry guys forgot to fill in my template!!
posted by You Can't Tip a Buick at 8:02 AM on May 8, 2015


I don't know what I expected since this happens in practically every social science thread (and in a bewildering number of natural science threads), but come on, guys, they define exactly what they're looking for and what metrics they use in the paper. Even if it doesn't fit your exact idiosyncratic definition of "troll", it's not really a "oooh, but on $conservative site you're called a troll but on $liberal site troll calls you" thing
posted by kagredon at 9:45 AM on May 8, 2015 [2 favorites]


They aren't actually all that specific, and there's a pretty rapid assumption that someone engaging in 'trolling' is, by definition, a 'troll.' Social sciences are complicated and nuanced, and the conclusions become less accurate the smaller your sample size gets.

Go back and read that paper from the perspective of a minority opinion you're sympathetic to. Not just 'liberal' vs. 'conservative' or something, but about a tangible subject that you personally would find important enough to go against the flow. Like, an otherwise reasonable forum has a group fondness for eugenics or something.

Back when Metafilter was a "don't read the comments" site for me, were the commenters who pushed back against the rampant sexism also 'trolls'? Like I said, I didn't read the comments here much at the time, and I certainly didn't participate. But it's a safe bet that some people did, right?

I've seen a lot of different online forums undergo changes in accepted group behaviors, and a lot of time, that involves some pretty disruptive patterns, including persistence and hostility and pile ons and almost always some inarticulate raging. And it's not just Metafilter that had the feminism conversion. That's been a very common one in a lot of general discussion forums, and it is always ugly.

I'm dead serious, BTW, so apart from obvious facile jokes and appeals to group ideals or whatever, explain, in concrete terms, the difference between 'a troll' and someone who is being piled on.

I am not saying there is no such thing as 'trolling' in this sense, and on those sort of wide-ranging comment systems that don't have a cohesive group identity, they are pretty easy to identify. However, on more "community" oriented sites, there's often a sort of smugness to groupthink, where forums will collectively settle on certain principles and get preemptively defensive when they're challenged.

And I mean, I have seen this happen over and over in small friendly social groups where I'm not the 'troll' in question. Where I agreed with and had played a part in those conclusions, but where even I could recognize that dissenters were being bullied. Just to address the obvious assumptions.

I realize that those characterizations of trolling make intuitive sense to a lot of people, but I suspect there's a lot of filling in going on behind that.
posted by ernielundquist at 10:54 AM on May 8, 2015 [2 favorites]


And it just occurred to me, concisely, that I am saying that this study is examining behaviors that are likely to cause someone to be banned.

The 'trolling' designation is an unnecessary intermediary step that adds no value beyond clickbait.
posted by ernielundquist at 11:02 AM on May 8, 2015 [2 favorites]


Do trolls actually serve a purpose?

the best ones do - they get people to reveal things about themselves that they would probably rather keep hidden - they trick people into exposing themselves as the irrational knee-jerks they really are - they get people to debate absurd things endlessly while they sit back and watch the show silently, having accomplished their mission

unfortunately there are very few high level trolls left - and you certainly won't catch them with the methods used in this study
posted by pyramid termite at 12:44 PM on May 8, 2015 [2 favorites]


Word.

As examples of old style Usenet trolling, the targets would often be newsgroups that were known to jump down newbies' throats for stupid shit. Star Trek and atheism newsgroups were prime examples where they'd really gang up on new people for really pointless reasons. By which I mean, not even saying something nasty or evil, but just some harmless ignorance that marks someone as out-group.

So a troll might go in and intentionally push their buttons to get them to go over the top. And if you built up slowly, piling absurdity on top of absurdity, you could get people who were totally in a froth over something downright surreal.

It can be very telling how casually some people assume that everyone else is implausibly stupid, and how cruel and hostile they can be to someone they honestly believe just doesn't understand something.
posted by ernielundquist at 1:49 PM on May 8, 2015 [2 favorites]


Also, http://adequacy.org/
posted by Drinky Die at 3:01 PM on May 8, 2015


"Einstein invented the telephone" was the first troll(ing) I saw on Usenet...20 years ago? Whoah.
posted by sidereal at 3:02 PM on May 8, 2015


also btw, not re: 'trolling', but fwiw...
A quick history of 4chan and the rightists who killed it
posted by kliuless at 4:44 PM on May 8, 2015


they trick people into exposing themselves as the irrational knee-jerks they really are - they get people to debate absurd things endlessly while they sit back and watch the show silently, having accomplished their mission

and

So a troll might go in and intentionally push their buttons to get them to go over the top. And if you built up slowly, piling absurdity on top of absurdity, you could get people who were totally in a froth over something downright surreal.

Somehow I'm still not seeing the benefit.
posted by tychotesla at 5:09 PM on May 8, 2015 [4 favorites]


Here's a classic, old-timey troll: "Ken M" on Horsey Surprise. I think he's pretty funny.
posted by Joe in Australia at 5:16 AM on May 9, 2015


It's fine if you don't really appreciate that type of trolling, and maybe it requires some familiarity with the culture it tends to happen in.

What isn't fine is conflating that type of trolling with the sort of shit where people harass grieving families and innocent people. You don't put the person who pretends they think Rush Limbaugh and the band Rush are the same guy and the person who threatens to come to that person's house and murder them for thinking that in the same category.
posted by ernielundquist at 10:47 AM on May 9, 2015 [2 favorites]


I don't troll here for a couple reasons. I think there was a pay barrier at one point, so that's a good step. The community and content here seems better moderated or pruned. The discussions seem more insightful and interesting overall. This seems like a nice place and I don't want to shit it up. On reddit, though, I really can't help myself. It's too easy, too rewarding, too consistent, and too fun to pass up and the entire place is already wrecked so it's hard to feel bad about stepping on eggshells when the chandeliers shattered on the ground. There it feels almost better to be trolling than otherwise, since even though you're obviously using the site, you aren't necessarily on the inside of the "club", which can be really embarrassing depending on much self-respect you have.
posted by GoblinHoney at 1:42 PM on May 9, 2015 [1 favorite]


The threaded nature of Reddit makes it a lot easier to go one on one with someone, for better or worse, without disrupting the discussion as a whole. You can feel better about messing with someone when you aren't shitting up the whole thread at the same time, which is what occurs in a flat format like Mefi.
posted by Drinky Die at 1:50 PM on May 9, 2015 [2 favorites]


anotherpanacea: "Thinking about this some more, I can't help wondering if all the algorithm does is determine the criteria moderators are using to ban people. So it's not necessarily the case that the people being banned are trolls, just that in the average community there's a list of behaviors that throw up red flags and get you booted."

The article seems to address that:
Very interestingly indeed, its performance didn’t suffer much if it was trained against normal below-the-line noise and then used on gamergate, or if it was trained against gamers and then used on libertarians (perhaps less of a surprise), or whatever. The authors argue that this is an indication that it’s picking up some kind of pondlife tao, an invariant essence of disruptive windbag.
posted by Lexica at 3:48 PM on May 19, 2015


It's the "invariant essence" that worries me. If there's a set of behaviors that get you banned regardless of content, that doesn't make that behavior trolling, it just makes it disruptive ceteris paribus.

But what happens when the ceteris isn't paribus? What happens, say, when you add a sign-up fee and have strong moderators? Then it seems like the essence varies.
posted by anotherpanacea at 6:20 AM on May 20, 2015


« Older "What are you guys doing, just sitting there...   |   Are we the only living thing in the entire... Newer »


This thread has been archived and is closed to new comments