19

Moderator from Ask Ubuntu here. Anybody else sick and tired of spam? We've had a lot this morning.

We've been through dozens of spam waves over the years but also several rounds of "improvements" to the automated systems. I love you guys but they just don't seem to be working.

I find three things particularly upsetting about all this:

  1. Many of the messages rely on easily identifiable common information (certain spam domains, keywords, phone numbers, etc) which are unique to spam.
  2. A community-run bot Smoke Detector is spotting spam 99.99999999% of the time with only very rare false positives.
  3. If I mark spam and that blocks the IP, that still leaves the messages from that account on other SE sites. Their moderators have to clean up.

And yes, the cycles of improvements might have brought us less spam, but it's still at thoroughly annoying levels.

I personally feel it's time to put some of the Stack Exchange spirit into this and crowd source it properly. You have a metric buttload of moderators and trusted users who know the system. They know what spam is.

So with that in mind I'm suggesting the following:

  • A single common blacklist for every site that contains entries that can contain simple single-line wildcards. I don't think we need regexes. And multi-line is dangerous.
  • Moderators (and/or super trusted users) from these sites should be able to suggest new block items that are applied to all sites.
  • It should take a few users from any site to edit the list, like a network-wide close vote.
  • Sites should have their own override whitelists that their mods and trusted users can edit.
  • On adding for consideration, existing posts inc. deleted that match the glob should be displayed so we can see the impact, network wide.
  • The content of the blocklist should be protected. So maybe moderator-only because we're already under an agreement.

I don't know what to do with the new posts that match. I've seen the opinions on hell-bans before but at the very least this should earn a number to flags so that it takes less work for our users to get this muck off our sites.

Improve this question
3
  • Presumably SE has some spam detection of their own... I know that I've seen posts of the percentage of posts that get stopped by the spam filter is a lot higher than I ever would have thought.
    – enderland
    Commented Jan 22, 2016 at 18:43
  • 1
    @enderland we do, and it keeps a lot of crap out of people's faces, but we need to finish the job.
    – user50049
    Commented Jan 22, 2016 at 19:47
  • 3
    @AwesomePoodles We don't really have that many false positives on the reasons that we're confident about. Thing is, we're not targeting 0% false positives, we're mostly targeting 0% true negatives. There are a few reason/site combinations that I'd be confident trusting it with spam flags on its own.
    – Undo
    Commented Jan 22, 2016 at 23:16

1 Answer 1

Reset to default
24

I'm kicking off an initiative to bulk-up the spam prevention layer of our shields. I designed the first one, and I'm not happy until spam is only something that people going fishing for it actually see.

Call it .. my own pet peeve, after being a moderator on Stack Overflow for a number of years. I hate it, and I really don't hate many things.

Specifically, I want to look more at what the community has come up with (we've been watching), and how we could incorporate it. I also want to nail down how mods could more easily evict vermin on a network level without asking them to go to chat at work and coordinate, in a way folks feel comfortable with. Stabs have been taken at both but stalled.

What we have keeps an amazing amount of crap from ever gracing or detracting from the attention folks give us, but we need more.

Challenge accepted.

Improve this answer
8
  • While I realize Oli didn't offer a challenge, it is in reality .. a bit of a challenge :)
    – user50049
    Commented Jan 22, 2016 at 19:42
  • 6
    Here's a real easy, cheap, improvement: In the API, have the shallow_user object return the user's Stack Exchange network ID (account_id). This would allow the existing tools, like Smoke Detector, to better catch cross-site spammers.
    – Awesome Poodles
    Commented Jan 22, 2016 at 22:00
  • 1
    I really respect the effort. I know you've been at this multiple times and I know several of the mechanisms you've implemented have stop a lot of rubbish. However, we had 12 spam accounts created this morning. Active on multiple sites. These 12 accounts have taken hundreds of flags to silence network wide. This duplication of effort still required to deal with such obvious (and easily-blacklistable) spam makes me want to cry. This should have taken a couple of mods a couple of minutes to nuke the lot.
    – Oli
    Commented Jan 25, 2016 at 11:45
  • 6
    Tim, when we see cross-site spammers, they typically post on 4 or more sites with as little as 26 seconds between posts. If you created a cross-site rate limit of at least 2 minutes between posts for low-rep users, it would really slow down these bots (and be a possible metric for more automatic action). Such a delay shouldn't harm legit users.
    – Awesome Poodles
    Commented Jan 27, 2016 at 5:57
  • 1
    Is there a reason why Smoke Detector could not feed posts into the low quality review queue? Likewise if a post on one site matches the spam filters the mods have setup on anther site.
    – Ian Ringrose
    Commented Feb 8, 2016 at 17:40
  • We're being battered with skin revitalisation junk today. Posts like this highlight how effective a score-based keyword blacklist could be (eg skin, revitalize, cream all score 2 points, botox, ageless and peptide (outside bio.se) all score 3 points. Any post scoring over 5 gets a bonus auto flag or two, anything over 10 gets thrown into liquid magma.
    – Oli
    Commented Feb 9, 2016 at 11:22
  • 4
    If you're looking for a low risk / high reward place to start, we (Charcoal / SmokeDetector) have found that just reporting things with a link at the end of the body yields 98.45% spam accuracy, with the vast majority of the rest NAA or low quality. Here's a list of what we're catching with it, and here's a list of the false positives. The fp's don't look very high quality.
    – Undo
    Commented Mar 7, 2016 at 17:31
  • There are a few restrictions we've put on that reason, ping me in TL or chat.stackexchange.com/rooms/11540/charcoal-hq for details. Still, blocking a few thousand spam posts over 6-ish months isn't bad.
    – Undo
    Commented Mar 7, 2016 at 17:43

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .