[ SmokeDetector | MS ] Bad keyword in body, body starts with title and ends in url, link at end of body, pattern-matching product name in body, pattern-matching website in body, +1 more: Makes it a lot more worth it because by Cons Hall on superuser.com
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, body starts with title and ends in url, link at end of body, +4 more: Keto Renew Diet pinto beans by nijiong12 on askubuntu.com
Q&A is a better format for discussions like this, certainly, given that you can comment on people's viewpoints directly instead of continuing in one long thread.
I think the only issue we have with Teams is that notifications aren't pushed out. Most people get notified either on GH or via email of new GH issues, but the same isn't true of Teams posts.
On the contrary, GH doesn't provide any sort of prioritisation, all comments are (relatively) the same, which means they all get (roughly) the same amount of attention/views
(back to GH/teams) I'm guily of only really reading the answers on teams. This is partially due to the fact that the post only really seems 'new' and worth looking at when there is a new answer , whereas new comments aren't as visible (and don't get voted on as much) (side note: do new comments get pushed to the front page? I don't know)
Sure, but it would've been hit by pattern-matching even if it wasn't blacklisted - and that would be okay, because it would be below AF thresholds. With a blacklist as well, it's an AF FP.
[ SmokeDetector | MS ] Blacklisted website in body, body starts with title and ends in url, link at end of body, pattern-matching website in body, potentially bad ns for domain in body, +3 more: Shakra Keto Diet by txuw laoho on drupal.SE
@tripleee That pattern looks like it's already caught by Mostly non-latin answer and Mostly non-latin body; append -force if you really want to do that.
tpu- by tripleee
@tripleee Added [a-z_]*(?:234_*)?7036517871[a-z_]* to watchlist
we could have a theory that pattern-matched websites should not be blacklisted; but this is opposite of what we usually want -- when we know something is bad, we want the "I know it's bad" reason, not the "heuristically probably bad" reason, even though their weights are actually kind of similar, roughly speaking
Sometimes the line even blurs, for instance I'd argue that (?<!paypal-)(customer|tech|technical)-?(support|service)\.com would fit into pattern matching too
so the semantics should be "blacklisted should exclude pattern match" which is a bit unobvious but not impossible to understand or implement
but I will say again what I have said many times before -- the problem with false positives from autoflagging are much easier to prevent by implementing stricter limits on autoflagging than by revamping the weight calculations
I think the line has blurred a fair bit over time due to changes in tooling - with the chat-based blacklisting, people are likely to add things like that to the blacklist simply because it's quicker & easier than editing the code itself
I think that this all comes down to the wrong things being blacklisted. If the blacklist is kept to things that are always, 100%, positively spam, then this would be a non-issue.
having expressions with uneven hit rates grouped under a rule which gets assigned a weight needlessly sacrifices accuracy ... we could calculate weight for every individual expression instead of lumping many expressions into the same weight just because they were added for the same reason
@ArtOfCode only just back from limbo but last time I checked, the majority of FPs were from users who had an autoflagging threshold significantly lower than the recommendation
if you recast this as percentage of FPs of all submitted autoflags you probably get a more meaningful number, 99.75% is from the total number of TP spam reports IIRC?
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, bad pattern in url body, pattern-matching product name in body, pattern-matching website in body, +2 more: Max Keto Burn Review by joeiskidmore on askubuntu.com
but if we observe that autoflagging FPs are a problem then let's at least explore what the current FPs/submitted autoflags percentage is, and decide where we want it to be
@tripleee That pattern looks like it's already caught by Pattern-matching website in body and Pattern-matching website in answer; append -force if you really want to do that.
But this is of course naive. Bias exists. Even if we're all spamfighting experts the reaction to flag without really looking on something that's already autoflagged is much higher
The problem is the bias is self amplifying when you know others have flagged as spam. To conventional users, spam flags look like downvotes. The mental leap from downvote -> flag as spam is higher then "There are already 5 spam flags on this"
[ SmokeDetector | MS ] Blacklisted website in body, body starts with title and ends in url, link at end of body, pattern-matching website in body, potentially bad ns for domain in body, +3 more: Shakra Keto Diet by txuw laoho on drupal.SE
> Pattern-matching website in body, pattern-matching website in title ---------- Title - Position 1-17: supplementgems.co Body - Position 1-17: supplementgems.co
90% human spam accuracy is lowballing too, it's probably a bit higher then that (which affects human chance to wrongly nuke on order of magnitude 5 higher)
indeed it is, but until we have decided on a way forward, do I stick to the current policy in Github or try to adapt to something closer to what you seem to be advocating?
it would help if the tags had better predictability, some tags have a prefix which helps group them but they obviously sprung into existence organically and without much thought
bear in mind also that we don't have unlimited volunteer time; either we need an automated process to fill these things in or someone(s) have to spend non-trivial time doing it manually
it doesn't really work that way though, the quick informal question is "do we have many posts which share this feature and is it hard to group them unless there is a tag for posts which have this feature?"
for instance a website isn't "reviewed" because I noted that it's hosted on wix or whatever. Ideally we'd still like to know what topic, which dns providers, etc.
though of course it's now also useful simply if you want to see which whois registrations use #domainsbyproxy (or a number of other anonymity services for which we have a tag)
but again I would love to be able to get rid of these tags if there was a way to do some sort of structured searching on the whois data (ideally live whois registrations, not our copy/pasted data from whenever somebody looked up the domain)
(but "ideally" here is really unrealistic the way things look now)
halflife could probably be persuaded to populate some of this data to reduce the amount of manual work
and some of the data which halflife currently produces would be nice to collect in a database somewhere (like IP addresses and ASNs, f'rinstance ... hard to come up with useful tags for those without producing a vast explosion in the number of tags)
@tripleee how would it be if I simply extended comments to be available on domains too? Halflife could essentially just comment the same stuff it puts in chat.
@ArtOfCode actually I was vaguely thinking the same, though no hurry as far as I don't expect to find time to overhaul and update Halflife properly any time soon
I have occasionally used the whois box for random free-form comments already
[ SmokeDetector | MS ] Blacklisted website in answer, pattern-matching website in answer, potentially bad ns for domain in answer, potentially bad keyword in answer: Subscribe to tags via app by Joe Johnson on meta.SE
[ SmokeDetector | MS ] Blacklisted website in body, link at end of body, pattern-matching product name in body, pattern-matching website in body, potentially bad ns for domain in body, +1 more: In this fight, you don't rally to move by Cheirt1962 on drupal.SE
[ SmokeDetector | MS ] Blacklisted website in body, link at end of body, pattern-matching website in body, potentially bad ns for domain in body, potentially bad keyword in body, +2 more: Shakra Keto Review by mxuw sroff on drupal.SE
@Makyen That pattern looks like it's already caught by Potentially bad keyword in body and Potentially bad keyword in answer; append -force if you really want to do that.
CI on 5e31ae6 succeeded. Message contains 'autopull', pulling...
@SmokeDetector Undisclosed affiliation, based on name match to a LinkedIn profile. User has posted recommendations to this API on two different SE sites (1, 2), under two different accounts.
@SmokeDetector I raised a custom mod-flag on both posts.
While I like the shout-out for us, that's not a good use of it for SE. A) It's false. Charcoal doesn't use Stack Overflow for Enterprise. We use Teams, which I understand to be a significantly different product.
B) If I was looking to use the product and saw the graphic, I'm much more likely to look up companies that I don't recognize (e.g. like Charcoal). Doing so would lead to determining that Charcoal isn't a large company (or a company at all), which would imply SE is cooking the information. overall it would result in me being much less likely to use the service, due to a lack of trust in SE's representation of it.
@quartata If they've re-branded Enterprise as including Teams, that's a bit better, but the words imply that the graphic is showing the "Fortune 500 leaders" that rely on Stack Overflow for Enterprise. Charcoal doesn't come remotely close to being in that category.
!!/report meta.stackexchange.com/questions/313359/… "test to see if Smokey can handle new form of full-form links to answers; see reported post for more info; please mark as false positive"
