Can Stack Exchange do a better job of detecting sockpuppets in this site?

Question

I have a strong suspicion that most if not all of the low-quality, bad-faith trolling questions about the Holocaust and Jews are done by the same person using multiple user accounts.

A lot of questions are about the same, or very similar topics. Of all the questions that I've helpfully flagged as being rude and abusive, many of them are about the Holocaust, and many of those that weren't instead were about Jews, or Nazi Germany.

A lot of the bad-faith questions about the Holocaust are similar in length: see the research in Should we increase the required question length of the quality filter to deal with Holocaust trolling?. This is despite the fact they involve different user accounts.

A lot of bad-faith questions about the Holocaust use similar phrases, telling other users to refute them, or saying that the Holocaust denial can't possibly be real (the person pretends to not be a Holocaust denier to increase the mockery), even though they involve different user accounts. Here's a sampling of these kinds of phrases:

https://skeptics.stackexchange.com/posts/45217/revisions

That can't be true. Refute this.

https://skeptics.stackexchange.com/posts/43439/revisions

This appears to be Fascist. Refute it. Do not allow these memes

https://skeptics.stackexchange.com/posts/42900/revisions

Refute this article please. Zionism depends on it.

https://skeptics.stackexchange.com/posts/42754/revisions

This must be true because the holocaust is real. Most certainly 6 million jews were gassed at Auschwitz starting in 1939. I just need to confirm this fact to make sure I do not appear a holocaust skeptic.

https://skeptics.stackexchange.com/posts/39608/revisions

Is this a holocaust denial website?

https://skeptics.stackexchange.com/posts/38322/revisions

I nean of course right. Why would there not be. Scientific evidence. Are witnesses scientific.

Did Soviets build a fake chimney in Poland?

This must be refuted, it cannot be allowed.

https://skeptics.stackexchange.com/posts/45513/revisions

Wonder if any of this is true.

Please let me know if you need clarification.

Questions that are almost exactly identical are posted by different users. For example, Did Soviets build a fake chimney in Poland? is by https://skeptics.stackexchange.com/users/52757/user52383 (why does 52757 not match user52383?), while https://skeptics.stackexchange.com/posts/45513/revisions was by user52603.

Users who have asked bad-faith questions about the Holocaust or about Jews might have a similar pattern of which Stack Exchanges they're members of. According to Limited in what I can do to investigate a suspicious question, the user who posted a question about Jews in the USSR was a member of Skeptics, Biology, Meta, Mi Yodeya, and Philosophy, while the user who posted a question about Auschwitz is a member of Skeptics, Biology, Meta, Mi Yodeya, Philosophy, and Politics.

The number of users involved in asking these questions is very large: there's user52308, user48066, https://skeptics.stackexchange.com/users/47060/a25bedc5-3d09-41b8-82fb-ea6c353d75ae, user45688, https://skeptics.stackexchange.com/users/28725/jacob-blaustein (note: this question involved anti-African-American bigotry, not anti-semitism), user42171, user40243, user40239 answering a question by user40238, user40193 and user40190 and user40188 answering a question by user40185, user37556 and https://skeptics.stackexchange.com/users/31326/d-j-sims asking questions that only differed in that one said "Is there a way to refute this image?" and the other said "Can you refute this image?", user37064, and so on. Unless we're under attack by a battalion of nazis, there has to be some sock-puppetry going on.

Is it possible for the Stack Exchange software to improve its detection of sockpuppets without causing damage to legitimate users?

I checked the meta stack exchange for sockpuppets, but most of them were about voting rings, rather than problematic people trying to evade bans.

Answer 1

I can't comment on a lot of this, not least to avoid giving ammunition to trolls.

I also have no desire to reward trolls by feeding them with any signs of outrage or even recognition.

Note that 7 of the 8 questions you cite had already been deleted, generally very quickly. (The 8th, from just yesterday, now is.) I think that shows we are on top of it.

Answer 2

I think we need to narrow down a well defined metric to actually determine if this is an issue we can reasonably solve (i.e. it's bigger than other issues, so worth the effort). On one hand we have Andrew Grim and tim listing decidedly holocaust trolling posts and highlighting how long they've been visible on the site, both saying it's too long so we should do something. On the other hand, we have Odd and T. Sar saying it's not really that long in their view.

We see the issue here. This metric by itself is subjective. If we give it full context, it will become objective.

The first obvious thing we need is a time frame to evaluate. Let's just say, for example, 2019 since the year is nearly closed. Next, we need the full tally of all the decidedly holocaust trolling posts (answers too if they exist). Then we need the time for each post that it was visible on the site (I think we should fixate only on deletion and neglect closure), and perhaps maybe the mean/median type stats would be useful. At this point we have a formula like:

There were X holocaust denial trolling posts in [time frame] visible for a total of Y minutes.

I feel like we can use this to at least gauge our progress, if we were to attempt extra measures for this issue above and beyond what we already do for all spam and trolling. There is one major issue: comparing time frames to measure progress like this assumes a constant site growth and user interaction.

If we hope to fully compare time frames, we need to also consider the site's total posts made in the time frame less other trolling/spam types. The issue here is that in theory, if the site grows by say a factor of 10 (10x the number of posts made per day), then the holocaust trolling would also 10x. The time frame cannot change, but the total minutes of visibility has no theoretical limit, so comparing a site of massive size to a small one with only the formula above would give the impression that the massive site has a major issue compared to the small one. So we need to include something like:

There were X holocaust denial trolling posts in [time frame] visible for a total of Y minutes.

There were A total posts made in that time frame, B of them were deleted, C of them were deleted for trolling/spam reasons, not quality reasons. Non-deleted posts were visible for D minutes in this time frame. All trolling/spam posts, C, were visible for E minutes.

X/A is our total holocaust trolling burden and X/C our relative holocaust trolling burden for this time frame. Y/D is our total holocaust trolling exposure and Y/E is our relative holocaust trolling exposure.

• we seek to have all four factors be 0 for the time frame given,
  • where X/A = 1 means that all posts were holocaust trolling,
  • where X/C = 1 means that all troll/spam posts were holocaust trolling,
  • where Y/D = 1 means that all posts were holocaust trolling and none were deleted,
  • where Y/E = 1 means that all troll/spam posts were holocaust trolling and none were deleted.

These metrics can reasonably be compared to other sites and time frames regardless of size. We can also use these metrics for other trolling/spamming types. We would just need to define that type well then tally them for the time frame, then replace variables X and Y accordingly.

If we can aggregate this data and compare time frames and different spam/trolling types, we might get an idea of an acceptable maximum toleration for our holocaust denial trolling. Without that data, it's premature to suggest acceptable maximums for X/A, X/C, Y/D, and Y/E.

---

Possible holes in the metrics so far:

• This assumes all spam and trolling has been discovered and dealt with.
• This only considers posts, not comments. Comments can be considered the same as posts, but perhaps leaving them out has merit, especially in light of the above issue. Also, only mods and staff can see deleted comments.