Slashdot Mirror
Crowdsourcing the Censors: A Contest
Before you get bored and click away: I'm proposing an algorithm for Facebook (and similar sites) to use to review "abuse reports" in a scalable and efficient manner, and I'm offering a total of $100 (or more) to the reader (or to some charity designated by them) who proposes the best improvement(s) or alternative(s) to the algorithm. We now proceed with your standard boilerplate introductory paragraph.
In his new book The Net Delusion: The Dark Side of Internet Freedom, Evgeny Morozov cites examples of Facebook users organizing campaigns to shut down particular groups or user account by filing phony complaints against them. One Hong-Kong-based Facebook group with over 80,000 members, formed to oppose the pro-Beijing Democratic Alliance for the Betterment and Progress of Hong-Kong, was shut down by opponents flagging the group as "abusive" on Facebook. In another incident, the Moroccan activist Kacem El Ghazzali found his Facebook group Youth for the Separation between Religion and Education deleted without explanation, and when he e-mailed Facebook to ask why, his personal Facebook profile got canned as well. Only after an international outcry did Facebook restore the group (but, oddly, not El Ghazzali's personal Facebook account), but they refused to explain the original removal; the most likely cause was a torrent of phony "complaints" from opponents. In both cases it seemed clear that the groups did not actually violate Facebook's Terms of Service, but the number of complaints presumably convinced either a software algorithm or an overworked human reviewer that something must have been inappropriate, and the forums were shut down. The Net Delusion also describes a group of conservative Saudi citizens calling themselves "Saudi Flagger" that coordinates filing en masse complaints against YouTube videos which criticize Islam or the Saudi royal family.
A large number of abuse reports against a single Facebook group or YouTube video probably has a good chance of triggering a takedown; with 2,000 employees managing 500 million users, Facebook surely doesn't have time to review every abuse report properly. About once a month I still get an email from Facebook with the subject "Facebook Warning" saying:
You have been sending harassing messages to other users. This is a violation of Facebook's Terms of Use. Among other things, messages that are hateful, threatening, or obscene are not allowed. Continued misuse of Facebook's features could result in your account being disabled.
I still have no idea what is triggering the "warnings"; the meanest thing I usually say on Facebook is to people who write to me asking for tech support (usually with the proxy sites to get on Facebook at school), when they say "It gives me an error", and I write back, "TELL ME THE ACTUAL ERROR MESSAGE THAT IT GIVES YOU!!" (Typical reply: "It gave me an error that it can't do it." If you work in tech support, I feel your pain.) I suspect the "abuse reports" are probably coming from parents who hack into their teenagers' accounts, see their teens corresponding with me about how to get on Facebook or YouTube at school, and decide to file an "abuse report" against my account just for the hell of it. If Facebook makes it that easy for a lone gunman to cause trouble with fake complaints, imagine how much trouble you can make with a well-coordinated mob.
But I think an algorithm could be implemented that would enable users to police for genuinely abusive content, without allowing hordes of vigilantes to get content removed that they simply don't like. Taking Facebook as an example, a simple change in the crowdsourcing algorithm could solve the whole problem: use the votes of users who are randomly selected by Facebook, rather than users who self-select by filing the abuse reports. This is similar to an algorithm I'd suggested for stopping vigilante campaigns from "burying" legitimate content on Digg (and indeed, stopping illegitimate self-promotion on Digg at the same time), and as an general algorithm for preventing good ideas from being lost in the glut of competing online content. But if phone "abuse reports" are also being used to squelch free speech in countries like China and Saudi Arabia, then the moral case for solving the problem is all that more compelling.
Here's how the algorithm would work: Facebook can ask some random fraction of their users, "Would you like to be a volunteer reviewer of abuse reports?" (Would you sign up? Come on. Wouldn't you be a little bit curious what sort of interesting stuff would be brought to your attention?) Wait until they've built up a roster of reviewers (say, 20,000). Then suppose Facebook receives an abuse report (or several abuse reports, whatever their threshold is) about a particular Facebook group. Facebook can then randomly select some subset of its volunteer reviewers, say, 100 of them. This is tiny as a proportion of the total number of reviewers (with a "jury" size of 100 and a "jury pool" of 20,000, a given reviewer has only a 1 in 200 chance of being called for "jury duty" for any particular complaint), but still large enough that the results are statistically significant. Tell them, "This is the content that users have been complaining about, and here is the reason that they say it violates our terms of service. Are these legitimate complaints, or not?" If the number of "Yes" votes exceeds some threshold, then the group gets shuttered.
It's much harder to cheat in this system, than in an "abuse report" system in which users simply band together and file phony abuse reports against a group until it gets taken down. If the 200 members of "Saudi Flagger" signed up as volunteer reviewers, then they would comprise only 1% of a jury pool of 20,000 users, and on average would only get one vote on a jury of 100. You'd have to organize such a large mob that your numbers would comprise a significant portion of the 20,000 volunteer reviewers, so that you would have a significant voting bloc in a given jury pool. (And my guess is that Facebook would have a lot more than 20,000 curious volunteers signed up as reviewers.) On the other hand, if someone creates a group with actual hateful content or built around a campaign of illegal harrassment, and the abuse reports start coming in until a jury vote is triggered, then a randomly selected jury of reviewers would probably cast enough "Yes" votes to validate the abuse reports.
Jurors could in fact be given three voting choices:
- "This group really is abusive" (i.e. the abuse reports were legitimate), or;
- "This group does not technically violate the Terms of Service, but the users who filed abuse reports were probably making an honest mistake" (perhaps a common choice for groups that support controversial causes, or that publish information about semi-private individuals); or
- "This group does not violate the TOS, and the abuse reports were bogus to begin with" (i.e. almost no reasonable person could have believed that the group really did violate the TOS, and the abuse reports were probably part of an organized campaign to get the group removed).
This strongly discourages users from organizing mob efforts against legitimate groups; if most of the jury ends up voting for the third choice, "This is an obviously legitimate group and the complaints were just an organized vigilante campaign", then the users who filed the complaints could have their own accounts penalized.
What I like about this algorithm is that the sizes and thresholds can be tweaked according to what you discover about the habits of the Facebook content reviewers. Suppose most volunteer reviewers turn out to be deadbeats who don't respond to "jury duty" when they're actually called upon to vote in an abuse report case. Fine — just increase the size of the jury, until the average number of users in a randomly convened jury who do respond, is large enough to be statistically significant. Or, suppose it turns out that people who sign up to review content to be deleted, are a more prudish bunch than average, and their votes tend to skew towards "delete it now!" in a way that is not representative of the general Facebook community. Fine — just raise the threshold for the percentage of "Yes" votes required to get content deleted. All that's required for the algorithm to work, is that content which clearly does violate the Terms of Service, gets more "Yes" votes on average than content that doesn't. Then make the jury size large enough that the voting results are statistically significant, so you can tell which side of the threshold you're on.
Another beneficial feature of the algorithm is that it's scaleable — there's no bottleneck of overworked reviewers at Facebook headquarters who have to review every decision. (They should probably review a random subset of the decisions to make sure the "juries" are getting what seems to be the right answer, but they don't have to check every one.) If Facebook doubles in size — and the amount of "abusive content" and the number of abuse reports doubles along with it — then as long as the pool of volunteers reviewers also doubles, each reviewer has no greater workload than they had before. But the workload of the abuse department at Facebook doesn't double.
Now, this algorithm ducks the question of how to handle "borderline" content. If a student creates a Facebook group called "MR. LANGAN IS A BUTT BRAIN," is that "harassment" or not? I would say no, but I'm not confident that a randomly selected pool of reviewers would agree. However, the point of this algorithm is to make sure that if content is posted on Facebook that almost nobody would reasonably agree is a violation of their Terms of Service, then a group of vigilantes can't get it removed by filing a torrent of abuse reports.
Also, this proposal can't do much about Facebook's Terms of Service being prudish to begin with. A Frenchman recently had his account suspended because he used a 19th-century oil painting of an artistic nude as his profile picture. Well, Facebook's TOS prohibits nudity -- not just sexual nudity, but all nudity, period. Even under my proposed algorithm, jurors would presumably have to be honest and vote that the painting did in fact violate Facebook's TOS, unless or until Facebook changes the rules. (For that matter, maybe this wasn't a case of prudishness anyway. I mean, we know it's "artistic" because it's more than 100 years old and it was painted in oils, right? Yeah, well check out the painting that the guy used as his profile picture. It presumably didn't help that the painting is so good that the Facebook censors probably thought it was a photograph.)
But notwithstanding these problems, this algorithm was the best trade-off I could come up with in terms of scalability and fairness. So here's the contest: Send me your best alternative, or best suggested improvement, or best fatal flaw in this proposal (even if you don't come up with something better, the discovery of a fatal flaw is still valuable) for a chance to win (a portion of) the $100 -- or, you can designate a charity to be the recipient of your winnings. Send your ideas to bennett at peacefire dot org and put "reporting" in the subject line. I reserve the right to split the prize between multiple winners, or to pay out more than the original $100 (or give winners the right to designate charitable donations totalling more than $100) if enough good points come in (or to pay out less than $100 if there's a real dearth of valid points, but there are enough brainiacs reading this that I think that's unlikely). In order for the contest not to detract from the discussion taking place in the comment threads, if more than one reader submits essentially the same idea, I'll give the credit to the first submitter -- so as you're sending me your idea, you can feel free to share it in the comment threads as well without worrying about someone re-submitting it and stealing a portion of your winnings. (If your submission is, "Bennett, your articles would be much shorter if you just state your conclusion, instead of also including a supporting argument and addressing possible objections", feel free to submit that just in the comment threads.)
In The Net Delusion, Morozov concludes his section on phony abuse reports by saying, "Good judgment, as it turns out, cannot be crowdsourced, if only because special interests always steer the process to suit their own objectives." I think he's right about the problems, but I disagree that they're unsolvable. I think my algorithm does in fact prevent "special interests" from "steering the process", but I'll pay to be convinced that I'm wrong. Today I'm just choosing the "winners" of the contest myself; maybe someday I'll crowdsource the decision by letting a randomly selected subset of users vote on the merits of each proposal... but I'm sure some of you are dying to tell me why that's a bad idea.
Don't rely on the cooperation of self-serving and outwardly evil companies to send your message.
I'll take my prize in zorkmids, thanks.
It's ALL dark !!
Can you name that tune ??
So he's crowd-sourcing the crowd-sourcing solution. One more level and we'll make a black hole!
they want you to snitch on others and if you can't do that then they make you snitch on your self.
Your idea doesn't take into account the number of people who would sign up, just so that they could hit "Abusive" at everything. You need levels of meta moderation for this to succeed.
Crowdsourcing moderation using a random sampling of users? What a novel idea. I think even better would be to then allow people to review those moderations to make sure the moderators don't abuse their power. Call it metamoderation if you will. You know, slashdot should really implement something like this.
Also, this idea is fucking stupid for abuse reports, since there needs to be a clear definition of abuse in the ToS, and it would be impossible for Facebook to ensure their crowdsourced moderators are following that definition. So my better suggestion would be for Facebook to spend some of its $100 billion to hire more fucking moderators. Where do I get my $100?
This sounds a lot like the slashdot moderation scheme...
For those who did not know, you can get the source code behind slashdot here
I'd be more likely to deputize to people who you find are more reliable (basically, trusted moderators chosen from your randomly-selected pool after reviewing their decisions). Your system assumes that most people will be reasonable. I think that is an inherently flawed assumption, including for the very situations listed above. You can't trust that only a minority will think you should remove something that is against the mainstream view.
The painting mentioned as a profile picture is Courbet's Origin of the World.
Probably best not check it out at work.
Although, of course, it is on the wall of a major gallery where anyone can see it.
Sigs are so 1990s. No way would I be seen dead with one.
Suppose an atheist created a page about islam in turkish language and islamist turks find this abusive - hey because the person denies Allah - and they create a mass campaign with abuse complaints.
Let's assume turkey is 99% muslim. (https://secure.wikimedia.org/wikipedia/en/wiki/Demographics_of_Turkey)
The random picked people would statistically fit to that group and they'd also approve the ban.
So, what is the trouble for?
Doing a good solution for handling abuse is going to take a lot more than $100.
If you let anyone with accounts click on an abuse button, you will get a situation like the guy who did David's Farm on Youtube -- a bunch of people making shill accounts to slander the guy (of all the people alleging things, there has not been one single proof or conviction ID shown) and flag every video made as abusive. It worked -- even though David Rock was one of the top grossing YouTubers a year ago, he got tossed out on his ear due to Google not really caring to validate all the complaints that the bots/shills made, even though all the complaints were bogus.
If you take a representative sample, same thing... the botters will create tons of accounts, and by random luck, one of the bots will be chosen. Then the same thing will happen -- people who dis some ruler get their accounts banned due to "abuse".
Slashdot's system works because Slashdot doesn't get mainstream attention. If it were a top tier site, there would be botters with mod points slamming anyone automatically who speak out on a topic, as well as people spamming their stuff on one end of the Firehose, while their bots mod it up on the other end.
This is actually very hard -- all it takes is someone determined enough to make a system to bypass CAPTCHAs, and perhaps have a range of disparate IPs to proxy through and they pretty much win.
Remove moderation altogether. Problem solved. Or is genuinely free speech something you're trying to avoid?
Fuck you in the head? Don't mind if I do.
It is jarring for me to realize that pages are being taken down because they merely *offend* others. These aren't kiddie porn or drug dealer pages, it's just people talking about stuff. They talk about their friends, their enemies, their schools, their governments. It's not all flowers and happiness. If they want real people on facebook, they need to realize that some people are going to say unpleasant things.
Maybe have a counter at the top of the pages that says "this page has received N complaints" but leave the content there so all can judge for themselves.
There's actually been a lot of research on this topic the last decade. No great solutions imo, but a lot of research. Most of it better than random=trustworthy. Here's the problem. Say you have N users and M are a fake mob. F=M/N is your ratio of fake users. Now select 0.01 of your users at random. F=0.01*M/0.01*N ... the ratio is the same, so the mob still works. Of course, I'm assuming you don't know which users are mob users. If you did, why would you bother with randomly selecting some? This isn't even a scholarly argument, it's just a mundanely obvious observation. Here's some actual research: http://scholar.google.com/scholar?q=reputation%20systems
Imagine if you weren't allowed to use roads because a bus company complained about your driving 3 times. --skunkpussy
I'm really hoping for a decentralized social networking system, where everything is not controlled by the Big Head.
We need some sort of model for reputation/trust, similar to a credit score. If there is no information about you, you get a low score because you are unknown. As you create a history of your actions, they should affect your score. If you flag a lot of content, and your flags aren't accurate, then your credit score should be lowered. That way, if you try to flag items in the future, it will carry less weight since you are an unreliable source.
Newgrounds had something like this.
..but I have to say it's ironic that you're posting about this algorithm on Slashdot, a site whose moderation system has incorporated the best of your ideas for years, and yet that doesn't seem to come up when you're asking for ideas.
I like the Slashdot system. Moderators are assigned points at times beyond their control, to prevent just the kind of abuses you mention. There's appropriate feedback control on how moderators behave. The job of moderating (and meta-moderating) is presented and appreciated in such a way that people actually do it. People are picked to do moderation in a reasonable way. The process is transparent, and the proof that it works is that the Slashdot comments you typically see are actually not horrible (usually) and sometimes are quite informative.
Let's not stir that bag of worms...
I have two algorithms, and I suggest that they are more valuable if used together, and indeed, if all three including your algorithm are used together.
(1) Identify "clumps" of users by who their friends are and by their viewing habits. Facebook has an app that will create a "distance graph," using a published algorithm. It is established that groups of users tend to "clump" and the clumps can be identified algorithmically. For example, for a given user, are there more connections back to the clump than there are to outside the clump? Another way to determine such a clump is by counting the number of loops back to the user. (A friends B friends C friends A.) Traditional correlation can be used to match viewing habits. This is probably improved by including a time factor in the each correlation term. For example, if two users watch the same video within 24 hours of each other this correlation term has more weight than if they were watched a week apart.
Now that you have identified a clump -- which you do not make public -- determine what fraction of the abuse reports come from one or a small number of clumps. That is very suspicious. Also apply an "complaint" factor to the clump as a whole. Clumps with high complaint factors (complain frequently) have their complaints de-weighted appropriately. Rather than "on-off" determinations (e.g. "banned"), use variable weightings.
In this way groups of like-minded users who try to push a specific agenda through abuse complaints would find their activities less and less effective. The more aggressive the clump, the less effective. And, the more the clump acts like a clump, the less effective.
(2) Use Wikipedia style "locking." There are a sequence of locks, from mild to extreme. Mild locks require complaining users to be in good standing and be a user for while for the complaint to count. Medium locks require a more detailed look, say, by your set of random reviewers. Extreme locks means that the item in question has been formally reviewed and the issue is closed. In addition, complaints filed against a locked ("valid") item hurt the credibility score of the complainer.
I hope this helps.
I will create a sig when innovation restarts in the U.S.
At its core, this sounds like a blend of Slashdot's moderation and meta-moderation processes.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
This is a much more difficult problem than it seems at first glance. Some other posters have already pointed out the problem of the "jury of your peers" concept with the example of the country Turkey. A similar problem arises if it is simply approached as "what is considered offensive in the host country" (in this case, the USA, since Facebook is based in the USA). Heck, there are pictures of my daughter in her soccer uniform that would be banned in Saudi Arabia because you can see her knees, never mind her ankles. Scandalous!
It is difficult to conform to all nations' "sensibilities" with regard to what is "inappropriate" without falling to the harshest restrictions, such as Sharia law or the Thai ban on any criticism of the Thai royal family.
Spotted Kangaroo (message 35830238 in this thread) has an interesting idea with using "trustworthy" members. I'm not sure how that "trustworthyness" would be calculated other than using a metamoderation system similar to Slashdot's. By using supposedly trustworthy members, and then allowing the Facebook staff to "metamoderate", especially in the instance of appeals against complaints, I think it could work reasonably well. It would take a while and considerable effort for shill accounts to build up enough "trustworthyness" to be able to have any impact since the shill accounts would have to show activity and not just longevity.
I like the "jury" system, though. It's better than letting people comment only on topics about which they have strong feelings. Given the large number of churches that use Facebook as the electronic bulletin board for their youth groups, I could see a disproportionate number of people moderating pro GLBT groups and pages down because it offends their beliefs. We need a random selection mechanism that still works fairly, such as trusting people to list languages understood honestly. I'd be useless in moderating a page in Turkish, for example.
Just a few thoughts. I hope that if someone notices a flaw in my reasoning that you could post a polite explaination of the flaw and propose a better solution. I'm not interested in the $100, so I thought I'd just toss a few ideas out for folks to use.
I like the idea, however your problem is you will always come across trolls on the internet, or people who just like screwing up systems. I would say this percentage on facebook is quite sizable, so i would propose these alterations(to be taken individually or all together or mix/match):
Assign a trustability value to each juror, that is hidden and modified in one of two ways(or both):
Have a pool of pre-existing cases(I'm sure facebook has tons of examples stored in their history banks).
In this situation facebook knows what the outcome should be according to their standards.
Have any prospective juror have a mix of "real" cases and these pre-existing cases mixed together for a trial period, say that first 20 cases they review have an unknown mix. This way they can't guess which ones are appropriate or not.
Use their verdicts on these existing cases to assign a juror a "reliability" factor on their verdicts on the non-example cases in their batch.
That way jurors who don't quite get the rules, or are causing problems, are easily weeded out and their vote counts less in the total verdict weight on their real cases.
Alternatively:
Trustability starts at 50%, so new jurors get half votes.
whenever a juror disagrees with the majority opinion by the polar opposite choice, lower their trustability rating.
Likewise when they are in the majority and it is not a middleground, increase their trustability.
Both of these improvements will lower the odds of troll or mob mentality, even if the control a decent size of the juror pool because their individual votes will be worth less, while being invisible enough to the end-user that they won't be able to tell they aren't being effective.
You never realize how much manually made unmanaged "linked" lists suck, till you have src.link.link.link.link...
Although you can't expect people to identify themselves as being knowledgeable about every conflict, argument, religious view, political wrangling or moral panic you could choose individuals from the same timezone and hemisphere that the complaints originate from (and maybe only ban the offending piece in that geography - unless more complaints are received from outside).
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
I like your idea of crowd-sourcing, and I came up with a few ideas while reading yours:
Test the judgement of the moderators.
When mods are called upon to moderate something make sure that they have no way of knowing if it's real or not. This way you can for example request of a mod if "MR. LANGAN IS A BUTT BRAIN," or any other previously non bogus post should be modded or not. This way, you can know before hand if the content should be modded and test the modding ability of the moderator.
Describe what how they should judge something.
When reporting bad content you can usually choose what kind of bad content it is, for example you could choose between:
copyright infringement
offensive
Advertising
(more things here)
It is important to have a distinction between the kinds of bad content because it allows you to both judge the users who mod and the judges. It could be possible that one user/mod can see the distinction between "offensive" and "non-offensive" content but not the difference between "copyright infringing" and "non-copyright infringing" posts. By having this system in place, you could judge each users/mods credibility.
This system allows you to have an overview of how credible each users complaint is and by also keeping track of how each moderators mods things, you could also select the right one for the job. This complements the first adjustment I mentioned and vice versa.
Don't just randomly select moderators
I'm pretty sure Facebook and YouTube keep track of a whole lot of things about their users. When things get reported you could try to search for unbiased mods based on all the data they collect. For example, one post could reported 100 times from one particular geographical location, you could then search for the mods from another location.
These are just the ideas I got off the top off my hat, maybe they'll be useful.
Lo and behold, for I am a sig!
I think you'd be hard-pressed to find a group of people who would familiarize themselves with the Facebook TOS well enough to actually enforce it. I'm afraid that what you'd actually get is a group of people who vote Offensive/Inoffensive based on whether they agree with whatever controversial topic is at hand. This puts any minority group (LGBT, religious organizations, etc.), as well as any controversial groups (pro-Life/pro-Choice, political groups, etc.) at a much higher risk than they are now. You need some sort of incentive for people to vote according to the rules, rather than voting for what they think is right.
You need to validate that the juror is a valid juror and not a random bot. For each juror, run them across a random subset of training cases (created by FB employees) inserted into their standard jury duty. If they vote the same way the people who made the the training cases for a high enough percentage of cases (90%), then they probably are valid people and you should count their votes when calculating whether to delete a group.
Something on the order of the number of hits to the number complaints. Lots of views and lots of complaints need more supervisory reviewers to make the call.
Simple.
Most people are censorious by nature to one extent or another. People tend to group together with like-minded individuals. Those two factors, plus a system that lets them trash those they dislike is an inherent recipe for disaster.
The solution, if there is one, is a simple filter process:
1. One complaint, per item.
2. Individual review of complaint.
3. If your complaint is blatantly unreasonable, you're banned.
Maybe if a member of "Saudi Flag" got permabanned the first time they filed a false report, things would change a little.
..but I have to say it's ironic that you're posting about this algorithm on Slashdot...
Given that even the summary says, "and could benefit from a crowd-sourced policing system similar to Slashdot's meta-moderation," how on earth do you think there's anyone involved that is not aware of /.'s moderation system?
The process is transparent, and the proof that it works is that the Slashdot comments you typically see are actually not horrible (usually) and sometimes are quite informative.
/. is not an awful system, not even bad, but there are some stupid design decisions. The comments you typically see were in the first few hundred, and most of them are karma whoring, that is, posts designed to get 3 or 4 people to mod them up in a kneejerk fashion.
What /. has going for it is that it churns through the news quickly, produces some reasonably good commentary, but it's by no means the standard to judge by.
| /dev/null
You can send my $100 there too.
make people participate in your whole system like stack overflow. Gradually, as they build up reputation, give them more power. Then if one of these guys does something odious, you can yank his/her priv's.
There is "prior art" for this idea, if you know where to look. OKCupid.com has has a crowd-sourced "flagmod" system for its Web site for years.
Require that abuse reports include a freeform description of why the suspected content violates the rules.
Then, don't ask the jurors whether the content violates the rules. Instead, ask them: Is (the freeform description) a true statement about the suspected content?
In other words: someone reports content for violation of TOS. Reason: "This content contains nudity."
The juror then gets the moderation request, and they answer a single question. Is: "This content contains nudity" true about the suspected content?
You're not judging the rules; you're judging the evidence.
Pretend there is some witty statement here.
When I look at slashdot and the way it gets moderated, I feel that either the culture of slashdot has changed a lot over the past decade or else I've changed a lot (it's sort of hard for me to tell objectively). I realize that communities and their biases are not a constant but there are a few topics where the slashdot moderation lately feels so alien to me that it has raised my internal astroturf alarm. Admittedly, I'm part of the problem for letting my mod points expire more often than I spend them, but I find when I force myself to spend them, I end up starting to moderate "+1 aligns with my biases" or "-1 I don't want to hear this point of view".
TL;DR
*Process is Irrelevant, Progress is Paramount*
As I bet many meta-moderation algorithms may use, use (double float) scores to gauge each users fitness to either report a site or moderate reports. People who report sites that are usually taken down increment their "credibility score". Contrary-wise, if they report sites that don't get taken down, their score goes down. Same for moderators. If a moderator makes a decision that usually agrees with the final verdict, increase their credibility. Moderators with a higher score contribute their score to the final reward pool of the decision thereby rewarding the "elite" moderators and those that agree. Same for reporting content. Reporters who report valid TOS violations gain credence as their decisions are supported. With enough iterations, a solid base of moderation is achieved. Control of the "elite moderators" group (or other users for that matter) can then be simplified to functions of their score(s). (This is all while randomly offering users to moderate for a specified (or random) amount of time to control "elite"-ed special interest groups) I would compare that to making users their own neural network of moderation.
This will never work. All "they" need to do is set up a network of Facebook "personas" like Aaron Barr was proposing (and the US Military was asking for, including the management software) and you can generate as many +1 or -1 as you want. For Facebook read Amazon accounts, Youtube accounts and also Slashdot accounts. No, they were not all coming from the same IP range or geographical area (as obtained through geolocation). Don't know if anybody already coined the term Socknet but that is exactly what it is.
Lets say the offending material is extremely and obviously offending. You want to send this image to 100 volunteers to be viewed and voted on? What If I have the my pictorial guide to gutting and skinning a pig uploaded to my profile. 1000 pictures... 100,000 volunteers just vomited a little in their throats. My account gets deleted, I create a new email address and I'm back! :D
Exposing volunteers to REAL abusive images/videos/text is IMHO worse then trying to catch false positives. If you want to be a forum moderator go for it, but to be a content moderator for an international site of millions of people is a bit like opening your soul to all the evil in the world.
User Accountability - Facebook users abuse reports should be confirmed on a random sample basis, every 10 abuse reports filed. If the abuse report is a fake report, the user should be dealt with on step based incident system. First incident, warning. Second-Third incident, temp suspension. Fourth incident, full account suspension.
This is a well-studied "Who watches the watchers?" web of trust type issue. While there is no perfect solution, there are a number of good approaches. This page on Advgato describes a good trust metric for reducing the impact of spam and malicious attacks. It wouldn't be that big of a deal for FaceBook to incorporate some such system. However, it would require FaceBook to actually care about about being fair to its users, which it doesn't. FaceBook exploits for financial gain the tribal desires of people to band together and be part of a group. So FaceBook's really uses its abuse policy as a way to force people to follow the rules of the bigger and more aggressive tribes. Such battles actually help FaceBook to be successful because it strengthens the tribal behaviors that benefit FaceBook's bottom line.
So all in all, no matter what brilliant, cost-effective, robust moderation/abuse system you design or crowd source, the very, very best that you can hope for is that somebody at FaceBook might pat you on the head and thank you for your efforts and say that they aren't interested in your contribution at this time.
So you want free content provided by your customers... but that lead to low quality content. So you wanted free moderation provided by your customers... but that lead to poor moderation. So you want free code and methodology to improve your moderation provided by your customers... at some point the obvious has to just smack you upside the head.
Why not "test" the jurors every so often to determine if they're really effective jurors?
It would work something like this: you would have a small group (employees of facebook, or wherever) that takes (actual) select complains and determine how their "ideal" juror would handle the complaint. feed these at random to the jury pool and if they're not voting the way they should, reduce (or remove) their voting power in effecting the outcome in the decision making process, alternatively if they have a strong history of voting exactly the way they should then their votes would carry more weight in non-test cases.
I wouldn't necessarily "kick out" jurors, but their voting power could be diminished to nothing if they have a very poor track record... I also don't think that the jurors should know that they're being tested nor, what their voting power is, nor that their voting power even has more or less weight than anyone else's.
Collector's Edition
How about a useful reply?
Totally agree that people like Morozov write off crowdsourcing without understanding it. One of the things that's fascinating to me is that crowdsourcing systems in general haven't learned from Slashdot's success with meta-moderation. Evaluating abuse reports seems like a great application.
He's helping kids circumvent security systems at their school to access banned sites and doesn't understand why he's getting complaints?
Here's your sign...
This reminds me of the video appeal in pro tennis. If your appeal was bogus you lose it, otherwise you get to appeal again. This makes you think twice before appealing, and will therefore reduce the review load.
Minor possible improvement: you get the right to enter a complaint only after some time since the creation of the account has passed, or if you have reached a certain amount of activity; this will deter the creation of shill accounts, or at least it will increase the friction/cost (time, energy) for doing so.
"In our tactical decisions, we are operating contrary to our strategic interest."
Your basic approach does seem to be vulnerable to someone registering a large number of "sleeper" accounts that wait to be called in to be a juror about something they care about (perhaps an upcoming attack). To help counter this: 1. An account can't be selected as a juror unless it's been active for a minimum amount of time, with actual activity. (Say, a month.) 2. Jurors which consistently ignore their "duty" get dropped from the list. [p] You would also want to attempt to weed out vandals from your juror list: 3. Jurors which consistently vote counter to the majority get dropped from the list. To handle borderline cases I would try: 4. For the crowd-sourcing system to function, a minimum of 2/3rds majority is required. 5. If a 2/3rds majority isn't achieved, then a paid moderator looks at the complaint.
The most fatal flaw in your algorithm is that you assume that the average user actually knows the Facebook TOS. They don't. I'm not sure it would necessarily be a bad thing, but most "jurors" would certainly end up just making a judgment based on their own values
No exceptional soul is exempt from a mixture of madness -- Aristotle
To further improve upon this I have a few suggestions. #1 Instead of the "Jury" making the final decision have it so that the jury is the initial buffer before the official complaint is registered for final review by facebook themselves. This should alleviate the amount of work they have to do and thus have more time to properly investigate the claims and the corresponding group/user etc. In this manner Facebook makes the official decision based on their ToS instead of randomly selected people and their interpretation of FB's ToS. #2 This suggestion most likely should be included with any option. Further improve the algorithm to select candidates that appear to have no biased opinions on the matter at hand. Eg a controversial group should not have a reviewer who is part of said group or has links to other groups either in favor or against the particular group in question. We know FB mines enough data to deliver adds based on a users preferences, hobbies etc. This could be worked into that. #3 If option 1 was to be thrown to the wayside have only "torn jury" decisions based on votes that are too close in numbers then sent to FB for final review. In this situation where a vote may be torn there is a high risk individuals either A) interpreting the ToS incorrectly or B) The randomly selected individuals may have a higher bias towards one vote or the other based on personal feelings and would allow FB (And various other sites if this type of review process was used elsewhere) to make a final decision based on the actual ToS instead of an interpretation
I think your system would work quite well. I don't think even a 100 jurors are needed. I would suggest an initial pool of 20. Now within that 20 of 17 say A, and 3 say B, go with A. If it's closer, say, 8-12, then enlarge the pool to 50. If it's still close expand the pool again.
One of the keys is that people should NOT be able to volunteer for Jury duty. This will keep the deck from being stacked by the self righteous.
I would suggest that Jurors be selected from people who have:
* Active accounts -- they are spending considerable time each day on FB
* Members of multiple categories of groups. E.g. They have broad interests.
One of the criteria for abuse reports: The weight of an abuse report varies with the age and activity of the account posting it. Thus creating a new account just for the action of making an abuse report is ineffective. Similarly, making a report from an account that is moribund is ineffective. So for each account, there is 'account status/repuation'
There should be a consequence of sending falsely abusive reports.
E.g. If you have sent in 3 abuse reports that are subsequently over turned, then with each abuse report, your account loses status. Essentially you are marked as someone who chronically complains.
For abuse reports that are essentially libel (there is no semblance between the complaint and reality) the accuser loses his account for a period of time.
to enable this feature, one of the choices the juror can make is
"The complaint has no basis in fact, and appears motivated by malice. The accuser should be censured."
Third Career: Tree Farmer Second Career: Computer Geek First Career: Teacher, Outdoor Instructor, Photographer.
You cheap bastards.
The jury's decision to remove an item should not be final. It should be a site employee with the final say.
Don't remove content in the first place. Put up a "this contains potentially offensive content" warning and let people click through if they want.
Once a complainer reports a link/image as offensive, remove it from their access so they don't see it any more.
bah.
I won't bother to echo everyone else's comments about how similar your system is to slashdot's mod/metamod system, but I did want to make one other point:
What's going to happen with a system like yours is that users will collectively bend and break the TOS, molding into the average of the userbase would prefer the TOS to be. This system will *not* enforce the TOS, even if the instructions say to do so. People will reference the TOS as a starting point, but ultimately the results of the abuse system's actions will be a new, organic TOS collectively designed by the userbase. FB has three basic choices once they realize this is happening: (1) Ignore it, and eventually the TOS will just become meaningless as the community moves on to having their own discussions about the effective TOS. (2) Try to monitor what the effective TOS is in practice and update the real TOS to match. (3) Just drop the TOS completely and state upfront that the abuse system is based on an organic, evolving community consensus rather than a pre-defined set of rules.
One method of pre-screening potential jurors might be to process their own published Facebook content online and analyse it for irrational thoughts or extreme positions on related topics to what is being voted on. Lets suppose you are looking at a potential take down of someone's web profile because of it being accused of being a 'sadistic cult'. What you would not want is to pick is a 'stacked deck' jury of another but opposite extreme like a religious faith, because that group in general would be more likely to be charged with emotion, and be less likely to make a rational choice. To have a sound judgement without personal emotional entanglement to a subject you would be better off selecting jurors without vocal religious connections or URL's pointing to any cult-like sites. So, how might this work? You create a directed graph of all key words pertaining to emotional phrases (in this case cults or religion), and then process the potential jurors 'published content' to see how highly they associate with both sadistic cults, and any particular religious idealism, and exclude those two general populations from any potential candidates. Choose the ones with the lower correlation to the subject. as they would likely have a more logical reason for voting yes/no. The down side is that experts in certain topics might also become excluded candidates where they might have actually shed some light on the case being voted on. What words are actually indexed in the graph for each topic and used for this kind of selection would have to be thought out very carefully. After all, those with the strongest opinions are not likely going to be happy with their role of sitting on the side lines.
Another but more long term method of ranking jurors would be to rank each participating juror based on their previous selection/votes in light of the actual outcome of that previous vote. If a juror votes contrary to the 'final accepted outcome' then that juror is ranked towards the outside the bell curve of what is "average" for that particular round, and would therefore be less likely to be selected for a similar topic. Someone that votes inconsistent is less likely to make the correct choice the next time, but they needn't be completely out of the running for a completely different topic. Eventually this juror ranking system would exclude those that are not serious about their role in the jury system and they would therefore not be selected again. Since the actual vote is using statistical averaging one bad apple won't spoil the system for any particular vote, and the system will become self correcting for future votes by performing careful selection of the proper jurors based on topic material.
I still believe that the best solution is to leave the censorship to professional mods, who at lest know what's really forbidden and what's not.
If you accept my assumption then the answer to your question lies in pattern analysis. If the people reporting are very tightly clustered, as in being friends with one another, having the same interests, liking the same pages or belonging to the same groups then the likelihood of mob behavior increases. The relevancy of such analysis can be determined by looking at the previous cases.
Now in order to establish if a page is merely controversial you have to look at the number of followers. A high profile page on a controversial issue will most likely generate a lot of reports, quite often from scattered groups of people.
Basically this analysis should generate some tags like "Possible mob behavior" or "Controversial issue" which would warrant more care from the mods.
right...