How can we empower per-site mods to more efficiently dispose of network spammers?

Question

Here's the problem, in a nutshell:

Glancy T. Pony is a mod on our mayonnaise site, and wakes up in the afternoon to find that her site has been littered by Miracle Whip astroturfers. As Glancy cleans up the mess, she notices that many of these spammers have accounts on multiple sites, where they've been dropping heathen sandwich spread everywhere.

Glancy, sadly, can only take action on her site, so she spends even more of her valuable time rallying help in chat to get the rest of the accounts nuked. She feels like there must be some way for her actions to do more, since it's so blatant.

I want to try to make this a reality, and I'm not talking about a site for mayonnaise (it's on-topic at Seasoned Advice). I'd like for per-site destructions of spammers to count more meaningfully when it comes to cleaning up network spam as a whole.

Here's what's blocking it:

• Mods are elected on a per-site basis, by individual communities. Are programmers willing to trust car mechanics and physicists to help keep Stack Overflow clean? (Along with the reverse)

• We've worked, quite extensively, at coming up with criteria for the system to notice and take action upon. E.g. if an account has more than X profiles destroyed for spam, and all profiles are eligible (less than 50 rep on any profile), automatically destroy them all. This works, but it turns out to be almost as slow as just waiting for the rest of the mods to notice in many cases. How can this be better, safely?

• While I am certain it would never come to pass, this sets up a system that if abused, could have some kind of awful consequences. We trust every single one of our moderators completely and extensively. It's just a possibility, however remote, that needs to be mentioned. What other guard rails could we put around it?

Please think about it. If you have ideas on how this could work, strong feelings on if it should or shouldn't be implemented, or just feedback in general - we'd love to hear it.

Part of my area of speciality at SE has been spam / abuse mitigation, and this keeps coming up. My response thus far has been it's not a horrible idea, let's see if a time comes when we really need it.

It's time to talk about it.

Answer 1

If I as a mod on Stack Overflow destroy an account for spam, you could immediately put a brake onto any linked accounts to require either more "I am human" validation or just hobble to one post per week and then flag up that account to the site moderators as one that's posted spam elsewhere. You could also feed the spam posts into the anti-spam program as more data (if you're not already doing that).

Then when the other moderators come online they'll see these flags and be able to act and hopefully the hobbling etc. will prevent too much spam being spewed onto the system in the meantime.

If they have accounts destroyed on multiple sites then intensify the hobbling or if enough accounts (more than half) are destroyed then automatically destroy the rest.

Answer 2

When we get spam, it is almost invariably by users with 1 rep point whose network profiles are also at or around 1 rep. I am reasonably certain that all spammers I've deleted should have been deleted network-wide.

This would suggest that it may well be a good idea to propagate deletions across multiple sites. At least as long as those users have very low reputation. Perhaps even under 10. I would also not object to mods from other sites having such powers in the communities I represent.

However, to avoid complications and protect against human error, we could always simply raise automatic flags. For example, I can imagine a system where

1. I destroy a user choosing spam as the reason.
2. An automatic flag is raised on that user on all sites where they have a profile.
3. In the flag review page, the mods of the target site get the option to delete the user or not.

This has the advantage that it greatly facilitates communication between different sites and bona fide spammers will be destroyed more efficiently while at the same time not allowing a mistake I might make to affect other sites. It sounds like the best of both worlds.

Answer 3

Create a "Destroy as spam account" function for moderators.

As part of the procedure do the following:

• Automatically generate a flag on any site that account is registered on for that user ("Network Spammer Detected")
• Automatically generate a flag for any other accounts with the same IP ("Possible Network Spammer")
• Prevent creation of new accounts on all of SE for accounts with that IP if they are "Destroyed as spam" on more than 1 SE site

This lets each site community handle their spammer(s) individually on a case by case basis.

But it still creates visibility for when network spammers exist on other sites which also affect your site.

Note that while this is going to "allow" a spammer to be destroyed on 1 site and then create another account, the third point will prevent mean that any spammer IP being destroyed on 2 sites will stop creation - period. Regardless of when the accounts were created.

---

Another thing which would help this:

• Automatically raise the second flag ("Possible Spammer") whenever a post has a community deleted spam post.

Right now, this is invisible to moderators. As a moderator I don't see users which post 1 thing and get community spam-flag deleted - providing visibility into this situation would help, too.

Answer 4

Blatant spam is the one universal issue that I would expect every moderator from every site to recognize. It does not need any domain knowledge or familiarity with the site to act on it.

One thing I think would be nice in general, and would avoid honest mistakes would be to actually mention the consequences of the destroy action in the dialog. Currently this dialog only mentions why the user should be destroyed, it might not be obvious to all mods that this has further consequences beyond the removal of the account and the posts.

While there are no differences in power between mods right now, I think this kind of tool could behave a bit differently based on seniority of the moderator and the amount of spam flags and user destructions the moderator has done. I'm mostly thinking about not taking the first few destructions from a new mod into account, just to avoid bigger consequences in case they simply misunderstood the tools.

I think in terms of signal, two destructions by (not completely new) moderators on two different sites should be enough to trigger the removal of all connected accounts. With the safeguard that there should be no significant real participation (upvoted posts) by those accounts.

Most spammers don't have any upvoted posts, I'd use something like the following criteria to restrict the tool:

• No more than 1-2 upvoted posts. I'd maybe even go with 0, I generally don't see upvoted spam, but I've heard it exists.
• No older, unflagged posts. If it's been here for a week and isn't removed, it probably isn't spam and the user at least not a completely obvious spammer.

These restrictions would ensure that even if the tool is triggered in error, the damage would be very limited. Those criteria are certainly exploitable by spammers, but you can always change them if that turns out to actually happen.

I'd strongly support such a tool as I think the current way of handling cross-site spammers is unnecessarily noisy. Pinging other mods in chat or CMs just to clean up some spammer accounts that might be abandoned or not is just not that efficient.

Answer 5

I suggest a networked moderator review queue, only accessible by moderators.
(Bear with me..it has good potential)

TL;DR;

Have a new review queue accessible by moderators only, which would be shown to all moderators across all sites.
When a user is marked as spammy, the system would raise a flag, add a new review in the moderator-only review queue, and then all moderators on all sites could see it and vote on what action to take (or add to a current review if user is already being reviewed).

Spam is not really site specific, so is ok for mods to vote on spam related things from other sites, imo.

---

Responses to your criteria (Tim)

Are programmers willing to trust car mechanics and physicists to help keep Stack Overflow clean? (Along with the reverse)

I don't think "spam" is site specific, however a review queue helps with this by requiring more than one vote to action something.

if an account has more than X profiles destroyed for spam, and all profiles are eligible (less than 50 rep on any profile), automatically destroy them all.

This can easily be worked into a moderator review queue, and an auto action taken once vote thresholds are reached.

While I am certain it would never come to pass, this sets up a system that if abused, could have some kind of awful consequences. We trust every single one of our moderators completely and extensively. It's just a possibility, however remote, that needs to be mentioned. What other guard rails could we put around it?

A review queue with multiple votes caters for this too.
And moderator decisions can be seen by all other mods, so anyone can nudge another mod and advise on their not-so-perfect decisions - "Hey, FYI, that was obvious spam because XYZ".

---

The basic setup

Again, this is all only accessible by moderators, and not even readable by non-mods.

Every Stack site would have a new "moderator review queue".

When a user is moderated for spammy actions by a mod on one site, the system raises a flag, and details of the actions would go into the moderator review queue on all sites.

If a flag has already been raised for that user, then there would already be a review in the queue, in that case this new flag would just be added to the current review of that user as additional data.

This would allow all mods on all sites to see current spam actions being made on other sites, by other mods. Moderators can then check the user account on their site and perform an action in that same (global) review.

It allows mods from any/all sites to cast votes as to how best to deal with this user, and not just based on their own site and own findings. This makes it a fast process, and more than a single person making a decision.

When there's a review needing attention (i.e. without an outcome) there would be a notification on the top-bar (like the "suggested edits notification").
This would cater for the fast action requirement, especially as any mod on any site can cast a review vote, so an outcome would come quickly.

The review queue should be updated from various places where a mod deals with a spammy user. e.g. A mod flags a user as spammy in an question or answer, the system adds it to the review (or a new review opened if none exists).
This stops mods having to manually raise or update a review - it's all just auto fed in based on their actions, and other mods from all sites can then review it.

The review

Various useful data of the user would be in the review, such as:

• Links to all spam posts flagged by all mods/sites for the user
• Other "potential/suspected" accounts that user might have
• Previous moderation actions against the user
• A link in the review to the user's profile - but the link would be different for each mod, and link to the site they are a mod for. Allowing mods to quickly check if that user is spamming on their site (instead of having to search for that user etc).

This allows a single place for all data to be studied about that user - instead of having to check other sites, chat to other mods, write stuff down, etc.

Review actions

Stack would set this up, and it's going to require more in-depth functions and consideration than I care to delve into here.
But for an example to illustrate how my idea would pan out:

1. Mod on SiteA flags spam from a user
2. System places data in the moderator review queue, all mods on all sites can see it and are notified on the top-bar
3. Mods on SiteB see the user has also spammed on their site (SiteB), they flag it and so it gets added to the review for that user (same review). This updates the review to show the user has also spammed on SiteB for all other mods on all other sites to see. A picture is building up
4. Other actions can be performed, such as any mods voting to ban user, suspend user, block account on one, some, or all sites, etc.

Or, point 3 might be non-existent, and user is only spamming on one site, so mods decide to vote to ban/suspend the user on that one site only.

The outcome will be determined by usual review queue criteria:
X votes by mods for Z reason

For example 5 mods vote "ban user site wide" and so the action is completed.

---

Complex review

For it to be useful, the moderator review won't be exactly like our current normal user review queues, where simply one action per user only, and when threshold met review is completed.

A moderator review should show various data about a flagged user, so a more concise and accurate decision can be made.
Such as the posts and sites the user has spammed on, and any suspected other accounts the user has.

So, while data is being gathered, new things come to light, so votes could change.

(These are moderators, so I expect that they won't just stupidly change votes for no good reason... or at least shouldn't.)

For example, changing one's vote:

1. UserA has a review up against them for spamming on one site
2. Mods start to vote "ban user on that one site"
3. A mod from another site flags spam on their site, which is fed into the review. Now it shows the user is spamming on two sites
4. Mods can then change their vote from "ban on one site" to "ban on two sites..or network wide" because more data is found

If votes are cast and review complete, and more data comes to light, mods want to change the outcome (re-vote).

There are two ways to tackle this. When a mod flags the newly found spam (from the same user but not previously flagged) the new data will either:

1. Resurrect the old review and reset the votes, so mods can vote including with the new data
2. Or a new review can be made, with a link to the previous review so previous data can be seen

There are pros and cons to both. I'm sure Stack would resolve this easily, perhaps with an option to re-open an old review if a user is seen to be "up to old tricks" etc.

---

I also suggest that different data within the review can be voted on separately.
This is because it's not going to be black and white "user has X spam posts - what should we do".
There will be other info and data to take into consideration.

For example:

1. Mods suspect the user of having multiple accounts, so the review shows "potential/suspected other accounts for this user" (and a list of the accounts)
2. Mods can investigate and gather data to confirm if the user has multiple accounts
3. A mod finds one of those accounts is the same user, so marks the "potential other account" as "confirmed" in the review queue
4. Other mods can see this, and vote accordingly, or change their previous vote cast

This too will change voting, because a user with multiple bad accounts should perhaps be banned completely, whereas a user with one account and spammed on one site, might be simply "not knowing the rules" (etc).

---

Review done

The review will be done and moved out of the queue when all tasks are completed.

For example, if the voting is complete for banning, and the user is banned, but there are still some unconfirmed "potential/suspect other accounts", the review would stay in the queue until dealt with.

This is because those other accounts might not be yet banned, and it could be that same spammy user who is banned on one account. The entire task needs to be completed.

---

Advantages

1. Fast outcomes - Any active mods from any site can jump in and action the user
2. No one mod can invoke a network wide ban, takes multiple "opinions"
3. No one mod can ban a user from a site they are not a mod on, takes multiple mods from various sites
4. While it would be new code, it will allow certain code re-use, and hooks into the current systems and setup (rather than entire new system)
5. Staff and Stack devs have great knowledge of review queues now, with tried and tested things, so you could fairly quickly set this up to what works best (rather than entire new system)
6. A users previous moderation will be stored for future reference. The system can pull out data to put in new reviews for the same user
7. No need for "clunky" chat to tell each other about stuff, make discuss, and a unanimous decision is made easily and without discussions
8. There is a historical record of spam moderation, which means new moderators, or those not so confident (small sites) can study the previous decisions and learn good practices. Without having to chat to or ask other mods
9. An outcome being from multiple moderators is more likely to make a better decision

---

More than just spam

This will take a fair bit of work, even re-using some current review code and knowledge of how reviewing works best (etc).

However, once this is up and running, tweaked, and bugs ironed out, other moderation things could be fed into here.

Not being a mod, I cannot suggest anything which would be useful, but there must surely be other things would would benefit a moderator-only review queue?

Answer 6

Suspend the network user for a short time (a day or so) so there is a chance to do a 'review' of the account. (Was it a single time mistake, or is there more)

Then the SE Community Managers, or other moderators (on the same site or other sites) can review the account.

If a user gets 3 or more 'delete votes' against him, BANG! Delete the account.

Answer 7

One of the important things I think per-site mods should have possibility to vote on nuking a network profile of spammer.

How can it work

• One mod on one site can vote to nuke a network spammer's profile, and only one per site

• There would be three votes from three moderators of three different sites required to nuke the spammer network-wide; apart from that at least one moderator, who has their site attacked by spammer must vote, the other moderators can discuss whether to vote or not in the room number four.

• One spammer can be considered network-wide once he spam on three different stacks (three is probably for consistence there :).

This should ease destroying network spammers - I guess the decisions of three moderators of different sites can be trusted very well.

(Also, maybe some special item in the mod-dashboard can be added, annoyingly blinking when someone votes to nuke the network-wide spammer).