@ByteCommander protip: add two spaces after each sentence, then search for all full stops and format them to increase font size, add double spacing, etc. Then, take all images and graphs and make them larger than necessary. That should give you an easy page or two :)
It is the chat room where the community built spam detection bot is maintained. We work to keep it updated with new spam patterns and validate that what it finds is spam.
I've seen a link used in a couple spam answers, which I think have gotten past Smoke Detector. Can I ask for that link to be added to a list of some sort?
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, link at end of body, pattern-matching website in body: Rarete Face Cream by thailexuvu on meta.SE
[ 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: cleanserenewdenmark.com/perlelux/ by MonHaw on apple.SE
@tripleee That pattern looks like it's already caught by Pattern-matching website in body and Pattern-matching website in answer; use !!/blacklist-website-force if you really want to do that.
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, pattern-matching product name in body, pattern-matching website in body: Dxl Male Enhancement by badch on stackoverflow.com (@quartata)
[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body, link at end of body, pattern-matching website in body: so your ruler will by Booke1931 on askubuntu.com
@SmokeDetector blog in German but I don't see an order form
among the other posts on the site, I see one for Pantothen with order banners but clicking them doesn't achieve anything
the BellaVei post has working banners
no, wait, they are clickable but they don't seem to go anywhere useful (I end up at Google, though this might be related to the anonymizing proxy I'm using)
[ SmokeDetector | MS ] URL in title, blacklisted website in body, blacklisted website in title, pattern-matching website in body, pattern-matching website in title, +1 more: healthpurelives.com/stackt-360/ by user251903 on apple.SE
[ SmokeDetector | MS ] URL in title, blacklisted website in body, blacklisted website in title, pattern-matching website in body, pattern-matching website in title, +1 more: healthpurelives.com/stackt-360/ by juanita cisco on astronomy.SE
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, pattern-matching product name in body, pattern-matching website in body: Dxl Male Enhancement by badch on graphicdesign.SE
@tripleee That pattern looks like it's already caught by Pattern-matching website in answer and Pattern-matching website in body; use !!/blacklist-website-force if you really want to do that.
@TajyMany I guess so, except that I don't do any metasmoke development. (metasmoke casts autoflags, not Smokey itself). You'll need ArtOfCode or Undo for that.
@TajyMany it just needs to be callable from Smoke Detector, we can implement a rule which triggers when it says so, and then the regular rule weighting will apply
so just tell us what the API is and we can start working on it
@tripleee ah, that's an option as well. But we need to be careful how this is weighed, as this will increase the total weight of all posts (including FPs).
@tripleee That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; use !!/watch-force if you really want to do that.
Maybe I could use a Genetic Algorithm to automatically find a weight (which reaches the highest accuracy on 16,000 rows) for Smokey's output (which itself comes from weighted inputs) and DeepSmokey's output?
@Glorfindel huh? the rule will initially have zero weight; if it starts contributing usefully, we probably want to rescale the weight threshold, but that's a separate topic really
we have added new rules and kept the threshold, basically because there are opposing pressures -- accuracy improves (pushing the threshold down) but rules are occasionally added (pushing it upwards)
If we could get it done soon, I'd really appreciate it. I'd like to see if the FP/FN rate decreases a bit to include it in another presentation of mine in a few weeks :)
@TajyMany it seems to accept but ignore HTTPS ...?
also, the verdict is simply a text like <pre>Spam\n</pre>, could you return JSON instead?
@Glorfindel I'm not sure I understand how you mean, could be just my ignorance, but from what I can tell, the ML simply compares the text to previous training samples, so it's basically independent of any other rules we have (though of course there will be overlap when regexes trigger on the same features that the ML picked up)
Yes, I meant the overlap. We have that with other rules as well (e.g. pattern-matching vs. blacklisting websites) but that overlap is more limited, and less black-box. Don't get me wrong, I'd love to see some ML enhancing our capabilities and TajyMany is definitely on the right track, but this is not Just Another Rule so it should be treated with special care.
Especially if it only works on SO, it will be tricky/impossible to get the weighing right.
@YvetteColomb Anyone can suggest, but they go via an automatically created PR on GH which has to be accepted. You have to have the code admin role on MS in order for it to be instantly accepted.
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, pattern-matching website in body, pattern-matching website in title, +2 more: slimatrexnorway.com/perlelux/ by bobbysuarez on graphicdesign.SE
Anyone know of a good Review Page refreshing userscript? Annoying to have to reload it all the time to check if there's anything to review. (This is on Pets.SE btw)
