[ SmokeDetector | MS ] URL in title, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: supplementlab.org/skin-tek/ by Groy on graphicdesign.stackexchange.com
[ SmokeDetector | MS ] URL in title, link following arrow in body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: elevategffacts.com/bella-serata/ by gawemfxocom on puzzling.stackexchange.com
[ SmokeDetector | MS ] URL in title, bad keyword in body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: supplementsguide.org/slumber-pm/ by AntonRook on bricks.stackexchange.com
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body, pattern-matching website in title: boostupmuscles.com/power-testo-blast/ by grac786 on drupal.stackexchange.com
[ SmokeDetector | MS ] URL-only title, bad keyword in body, bad keyword in title, link at end of body, link following arrow in body, pattern-matching website in body, pattern-matching website in title: healthytipweb.com/garslimia-garcinia-cambogia/ by christopiue on askubuntu.com (@ThomasWard)
[ SmokeDetector | MS ] URL in title, bad keyword in body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: quicksupplementfact.com/slumber-pm/ by AntonRook on graphicdesign.stackexchange.com
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body, pattern-matching website in title: boostupmuscles.com/power-testo-blast/ by user69255 on drupal.stackexchange.com
could Metasmoke run on something with better connectivity? I don't gripe so much over the flagging latency but just opening a link to Metasmoke is slow business
@ArtOfCode I was going to have a look at that on the weekend, but if you get to it first that's great. It might get a little tricky though, as we have to reset the variable on both flag dialog closing and after feedback
The fact that you need to wait until the page is fully loaded, before you click the flag link isn't helping to, since the flag link also takes some time to load
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, link following arrow in body, pattern-matching website in body, repeated URL at end of long post: The Most Effective Skincare Product by harahboore on graphicdesign.stackexchange.com
@YvetteColomb yeah, it could also be a really sneaky spammer, but many of the questions look quite genuine (though typically VLQ) at least in isolation
@Glorfindel I marked it as a spammer. No other posts. And a gratuitous link to that shopping website. Unnecessary to get the specs on that electronic bit
@tripleee I think it's spam, based on first time question and that link was not necessary to describe the article. In fact you couldn't get it to be more "buy me" than that link
I think so. When I was in here day in and day out, I would pick up the same websites being used, in the guise of questions, and you start to see patterns in spam that people may not notice as a "one off" post.
that would be less prone to false positives based on individual keywords, and potentially identify content we are currently missing based on tangentially related keywords
@angussidney something like a bayesian classifier, it would pick up the words to look for via training
so you'd train it on previous posts and it would pick up whatever phrases the spammer used in the past
not necessarily just keywords for this particular topic, but (speculating) perhaps turns of phrases they just happen to like to use
depending of course on how you pick the features
trad spam filters just use "bag of words" on the individual words in a message or document
@tripleee I believe that in the past (before I arrived) we have tried doing bayesian classifiers, natural language processing and machine learining, however it was not as successful as our current manual tweaking
but there are various n-gram approaches which look for pairs of words in the same document, or (near-) adjacent word pairs, or whatever
@angussidney I'm thinking the problem with that is often that the categories are too relaxed, you categorize into "bad" and "good" and the filter can't find a good explanation for why a particular group of unrelated samples are bad
but you could fix that by training individual "testo", "weight loss" etc classifiers
[ SmokeDetector | MS ] Bad keyword in title, link at end of body, link following arrow in body, pattern-matching website in body, repeated URL at end of long post: Fights Dark Circles And Wrinkles by rommelinarrow on drupal.stackexchange.com
@trip If you want to try doing classification, go for it. It'd be awesome if we could get that right. Just be aware that at least two of us have tried it before and failed.
On an unrelated topic @ArtOfCode, I was thinking, when smokey breaks it usually breaks during spam hour when no-one is online except me and a few lurkers. I have a Pi sitting on my desk which I don't use much, so I could boot it up as a backup copy if needed, but AFAIK I need special keys to allow it to create posts on MS etc. If you and the other admins were to allow me, would it be possible for me to get a key?
@angussidney Bear in mind that I also have the backup script running on EC2, so if smokey dies, the script will bring it back up within a 10 minute maximum
@Magisch AFAIK, there are still requirements on what you have to pass from homeschooling, too
@angussidney Yeah, that wouldn't be a bad thing. I believe zaq got a key for that reason, but now that he's not around it's probably worth giving someone else one.
School standards here exist mainly to prevent kids from being mistaught or indoctrinated, and due to that teachers are audited and controlled and carefully selected and well paid.
@angussidney I... don't actually know. I think I'm hosting it at the moment to give Undo a chance to fix the Pi, since that had errors that were causing frequent crashes
[ SmokeDetector | MS ] URL-only title, bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: realcoloncleansingworks.com/geneticore-boost/ by Etszsloi on askubuntu.com (@ThomasWard)
There is also that Neal person who seems to be copping a bit of criticism due to his lack of activity. However, his high rep means that he will probably get a lot of drive-by voters.
I think it would be a good idea to hide all rep counts in election stage. That way it becomes more dependent on the info in the nominations and comments rather than the scorecard.
Heh. My lecturer just asked if I'm sure I'm a first-year not a third-year. I quote: "I've yet to meet a first-year who can actually write as many languages as they claim."
