SmokeDetector The Next Generation sounds like a movie title, not a spam-detecting bot. SmokeDetector Next Generation sounds... slightly more on the mark.
I've been working on a new reason, but it involves "a bit" of machine learning and processing time. The idea is to automatically predict if a domain we see is good/bad. It works decently well on my test cases...except for one domain.
That domain? Mine.
sighs loudly
Expected: good, Prediction: bad, Probability Good: 0.0311940564041, Probability Bad: 0.968805943596
@tripleee It's parsing two giant lists of good/bad domains. The good domains are domains that have been seen on SO on upvoted posts. The bad domains are anything MS has detected, plus domains provided by 8 spam blacklist services
Nothing fancy.
It does appear to be susceptible to poisoning though. For example, this post had a Wikipedia link. The wikipedia test was only 80% sure it is good
Expected: good, Prediction: good, Probability Good: 0.80882127861, Probability Bad: 0.19117872139
So, I probably need some whitelisting of a handful of domains, but now that I'm this far, I can at least do that.
@Andy I don't think you need it to be real-time, just have a reasonably recent dump (i.e. probably not more than a few years old) as a starting point for building a whitelist
@tripleee you wouldn't want to build a whitelist on all the sites in that page blindly.. #38 is a NSFW website which we surely wouldn't want to whitelist.
yeah, seems like a good starting point; but we would want to manually review each website
@tripleee it's not the NSFW point I was mainly addressing; but that website on a SE post will 99% be in a rude answer, so we wouldn't want to whitelist it
[ SmokeDetector | MS ] URL-only title, bad keyword in body, link at end of body, pattern-matching website in body, pattern-matching website in title, +1 more: bellecomplexnorway.com/rejuvanelle/ by Vzszsloi on askubuntu.com (@ThomasWard)
FR could the FIRE rendering change when you have already clicked on an item and flagged it? I find myself clicking a second time fairly often just to check whether I have already flagged something in messages more than a few minutes old
hovering over the feedback is distracting because you can't click there, it's easier to open the pseudo-dialog and dismiss it or flag if you haven't but it would be slicker still if I didn't have to click anything
the shuffling that occurs when you hover on a post is also somewhat distracting when you try to home in on one of the posts in a sequence, could the expand action be slightly delayed so it didn't happen by accident so easily when you try to zoom on the adjacent post in the transcript?
@Cerbrus Could you also add a feature that if you use the FireAPI to open a post report, and another report is already open, that that report will be replaced, I sometimes forget to close the actual fire dialog when flagging using my notifications script
@Ferrybig You could just call the close function regardless of if a popup is open
but I'll have a look at that later
@Ferrybig Actually, It's not that simple on FIRE's end...
@tripleee Not sure what shuffling you mean there
@tripleee I'm gonna write that down as a FR for now. flagging logic is a bit all over the place, no simple way to just apply the class to the fire button.
I was told to post in here about a spamming account. has 3 Q and 1 A in startups plus a new account in english languague, network account is at stackexchange.com/users/10676233/opaly.
@GypsySpellweaver flagging on the site (with no userscripts) reports the spam to Stack Exchange admins and mods but not to us volunteers or our MetaSmoke database
hmmm, depends on the site also, but that one definitely looks like it's promoting a particular product, though the brand and web site are not directly visible
could be that they hope people will google and find them, or they planned to add promotional links once they see whether the copy will remain on the site
[ SmokeDetector | MS ] Bad keyword in body, link following arrow in body, pattern-matching website in body, repeated URL at end of long post: HOW TO APPLY IT? by oualided on apple.SE
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body, repeated URL at end of long post: Locating The Skin Care by cottormic on askubuntu.com (@ThomasWard)
I wish the network profile page would auto update like posts do. Been camped on opaly's page watching for new account creations or posts on existing sites.
@Henders It's a two step process. 1/ flag the spam nukes the post and deletes it 2/ if user is spammer nuke the user. Step 2 is not automated although some mods use a userscript
Actually I'm not sure if a mod spam flag will report here using FDSC. The post may be deleted before FDSC can report it. Will test that next time I see a spam before smokey does.
@DavidPostill reporting via FDSC is broken at the moment, but once it's fixed, metasmoke/Smokey is too slow for that. You'll have to !!/report manually.
@GypsySpellweaver basically just pop in here and ask, there's the "whom to ping" link in the starred messages to the right to help you pick someone to watch for
You don't really need much time to contribute here anyway - most of us just keep a tab open here while at work or when we're doing something else, and we quickly flag some spam when any comes through, then go back to our work.
