10

I'm seeing multiple spam questions advertising "Keto" pop up on Meta on a daily basis, as you can see in this MetaSmoke search.

While it is true that MetaSmoke / Smokedetector does handle these questions quickly, It may be beneficial to update SE's built-in spam handling to catch these questions before they appear on the site.

That would stop that significant quantity of spam from appearing on the über-meta.

  • 8
    I'm not sure how the in-built spam handling works, but while there is a stack of Keto spam for example on chemistry the term is used a lot plus presumably a few other sites: chemistry.stackexchange.com/search?q=keto – PeterJ Dec 11 '19 at 11:02
  • Good point, @PeterJ. I'm pretty sure there are site-specific settings, but that's as much as I know about SE's spam detection. – Cerbrus Dec 11 '19 at 11:03
  • 8
    For what it's worth, Smoke Detector has what we call a watch on the keyword keto. That means we get a notification in Charcoal HQ, but it's not automatically flagged as spam. We cannot move this to a proper blacklist item because of the false positives on a few sites where ketones are frequently discussed (health, chemistry, a few more). Feel free to peruse the metasmoke search results ... – tripleee Dec 11 '19 at 11:15
  • 5
    ... (try also post titles; registered users can use regex search as well). For the record, the regex \bketo\b currently has 83 false positives on post bodies, predominantly from a few health and chemistry sites, but also a number of Stack Overflow questions where the code coincidentally handles data for nutrient sites or whatever. \bketo\b in question titles has 6967 true positives and only 14 false (!) – tripleee Dec 11 '19 at 11:15
  • 1
    I am curious of what are the spammers trying to achieve. SE is notorious for deleting spam at light speed, and surely the resources needed to create the spam are not free. Even if it isn't deleted in a few minutes, the odds of anyone clicking seem low as well. – Tomáš Zato - Reinstate Monica Dec 11 '19 at 16:51
15

That just becomes a game of whack-a-mole with permutations and regular expressions. As noted in comments, spammers very quickly adapt to changes in the blacklist.

Keto, for instance, becomes:

  • K3to
  • Ke10
  • K3t0

... and so on. Even with a lot of optimization, each entry in the spam bucket / bad word / bin list causes a little more work to be done anytime there's untrusted data to sanitize. While we can scope blocks so that certain character combinations can be restricted in titles or tags and more, it's still expensive.

So, any strategy where that's our first move just ends up creating this giant boneyard of mangled strings that will eventually result in a noticeable impact on performance. We do occasionally use the blacklist for this, but it's almost never as a first option, and is always temporary.

Smoke Detector does a great job of making these vanish very shortly after they appear, which is what matters most to keep things tidy and keep our weight in search engine indexes. The honeypots our spam system uses are also tremendously effective at stopping these things soon after they start (honeypots are questions that get more than a few spam suggested edits over a period of time, so the system actively holds long grudges against entire networks it catches making subsequent ones).

As long as you flag it as spam it'll be handled. While spam bucket checks does help stop the worst stuff before anyone has to see and identify it, most of the stuff spammers get paid by the piece to plant requires someone to see it and identify it at least once before we train on it, which has the side effect of a few remnants being left around.

  • 2
    Note: Right now it's "keto" but before that it was a famous psychic in India, and before that it was mail order romance, and before that it was another wave. We learned the hard way that blacklisting (1) shouldn't be your only defense and (2) isn't a strategy for keeping it out. – Tim Post Dec 11 '19 at 15:38
  • Okay, are there other (additional) measures SE can take? This is the first time I noticed the spam being this focused on one subject. – Cerbrus Dec 11 '19 at 15:40
  • 3
    @Cerbrus Yep, and I'm doing it now. Each site has settings on how it contributes to and reads / reacts from spam data, so I'm looking to see if cranking up the "grudge" levels will help. – Tim Post Dec 11 '19 at 15:41
  • Sweet, thanks for looking into it! – Cerbrus Dec 11 '19 at 15:42
  • 9
    I'm gonna beat that spam with my discount gucci purse. – Tim Post Dec 11 '19 at 15:42
  • 1
    I thought feeding blue pills to spam fixes the problem... – Luuklag Dec 11 '19 at 15:55
7

Keyword blacklists, IP blocks, and similar preventative measures will never keep up with the tenacity of spammers. And Smoke Detector works well. Really well. I hardly see spam flags anymore, and I pop onto my site a lot throughout the day for quick flag checks.

Just flag the post as spam. Don't edit it, don't downvote it, don't use another flag. Flag as spam and move on.

  • Yes, on Drupal. Drupal hasn't gotten close to as much spam as SE meta has recently, though. I don't think your experience with spam in general on a different site really is that relevant to this specific spam subject on SE meta. – Cerbrus Dec 11 '19 at 15:02
  • Now, when a significant portion of spam contains a certain keyword, it's an easy win to blacklist that keyword. Sure, eventually, they'll figure out that name is blacklisted, but have you ever seen vi4gr4 spam? Changing the product name doesn't really work for spammers. – Cerbrus Dec 11 '19 at 15:08
  • @Cerbrus Drupal Answers was a primary spam target in the pre Smoke Detector days (around the time when Ask Different and the Ubuntu sites were, too). I have probably handled thousands of spam flags, and would spend significant time of time each night destroying users from posts that caught six flags beforer a mod. This stopped as soon as Smoke Detector appeared. – mpdonadio Dec 11 '19 at 15:12
  • @mpdonadio That's true, I haven't really seen any spam posts on DA the last month. – user569408 Dec 11 '19 at 15:13
  • 1
    Yes, I know. SE / SO are also massively helped by Smoke Detector. I also mention that in my question. That said, this answer is completely ignores the fact that 50% of recent spam on SE meta is for that keyword, which would make it an extremely easy win for Stack Exchange's built-in spam blocker. Smoke Detector is great, but SD is an antibiotic. Ideally, we'd prevent the infection from occurring in the first place. That's what I'm requesting, and that's what this answer does not answer. – Cerbrus Dec 11 '19 at 15:15
  • 1
    @Cerbrus I would still argue that keyword blacklists aren't effective based of first hand experience. Spammers will adapt faster that the SE team can keep up with the blacklists. – mpdonadio Dec 11 '19 at 15:21
  • Then maybe the blacklist should be easier to maintain. – Cerbrus Dec 11 '19 at 15:26
  • 3
    The core of Smoke Detector is a collection of keyword blacklists. While your general observations (spammers switch product names, web sites, IP addresses, etc) the efficacy of Smoke Detector is proof that we can and do keep up with the top spammers on these sites. My concern is that we are perhaps too specialized on rapidly mutating pharma spam from India promoting various forms of quackery to Westerners. – tripleee Dec 11 '19 at 15:40
  • It's about time SE integrated (some aspects of) SD... Imagine how clean the sites would be... – Cerbrus Dec 11 '19 at 15:45
  • @Cerbrus A few years ago I proposed allowing diamond mods to handle spam flags network wide. With 600(?) mods, blatant spam on sites would last a few seconds. – mpdonadio Dec 11 '19 at 15:47
  • @mpdonadio: Using a global spam flag queue? That could be quite effective, if the list isn't too daunting. – Cerbrus Dec 11 '19 at 15:48
  • 1
    @Cerbrus Yeah, global queue, meta.stackexchange.com/questions/192881/… – mpdonadio Dec 11 '19 at 16:41

You must log in to answer this question.

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