« first day (1200 days earlier)      last day (2619 days later) » 

9:00 AM
in SO Close Vote Reviewers on Stack Overflow Chat, 8 secs ago, by M.A.R.
Rude/Abusive http://stackoverflow.com/a/42527308/5210517
 
@Cerbrus sorry, had to leave a comment ... don't be sad ...
 
Got another one for you: graphicdesign.stackexchange.com/questions/86002/… (now deleted)
 
[ SmokeDetector ] SmokeDetector started at rev 13aae02 (angussidney: Strip code blocks --autopull) (running on teward/Aurora)
Restart: API quota is 6011.
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, link at end of body, pattern-matching website in body: Big-Name Brand Skin Care Items by Galeeastman on drupal.SE
tpu- by FelixSFD
[ SmokeDetector | MS ] Few unique characters in answer, mostly non-Latin answer: How does JSON translation work? by 3.1415926535897932384626433832 on stackoverflow.com (@dorukayhan)
 
@SmokeDetector tpu-
 
@SmokeDetector K
 
9:05 AM
[ SmokeDetector | MS ] Blacklisted website in body, repeated URL at end of long post: Are there any known side-effects? by xtestesto on stackoverflow.com (@dorukayhan)
tpu- by Glorfindel
 
@SmokeDetector K
 
@rene I'm sad
 
[ SmokeDetector | MS ] Blacklisted website in answer: How to test android apps in a real device with Android Studio? by Daniel Jacob on stackoverflow.com (@dorukayhan)
 
@Cerbrus you're more then welcome ... ;)
 
That said, the same "we need tests before changing stuff" can be applied to normal changes as well...
 
9:09 AM
tpu- by rene
 
@SmokeDetector tpu?
I didn't notice anything suspicious
 
@Cerbrus yeah, that is true. First question could be: is there a unit test framework and if not, which one should it be and then write the first test.
@M.A.R. I did on the last link, that is riddled with ads
 
Shrug I'll trust you
 
@rene that would make all these posts spam as well ...
 
@Glorfindel aargh, I'm off flagging spam ...
or oops
@M.A.R. retracted my flag
 
9:15 AM
I did as well
 
clear from MS as well?
 
@SmokeDetector F
 
(never mind)
 
[ SmokeDetector | MS ] URL in title, bad keyword in body, pattern-matching website in title: gomusclebuilding.com/iron-bull-edge/ by kianthomas on meta.SE
tpu- by Glorfindel
[ SmokeDetector | MS ] Username similar to website in answer: Trying to resize the navbar, change font size, and center the text by examhelpline on stackoverflow.com (@dorukayhan)
tpu- by Yvette
 
@SmokeDetector K
@SmokeDetector K
 
9:25 AM
@tripleee you do realise that those two users are different, right?
Just with the same name
 
@angussidney uh, no, I thought it was the same guy
well obviously it is the same guy because he posted the same answer
 
The GameDev account isn't associated with the SO account either.
 
but thanks for looking into it
 
At least it wasn't when the spam was reported.
 
teach me to check first
 
9:26 AM
Also, I don't know whether or not it is possible, I seem to remember that we have been waiting for a FR for the API to be resolved so that we can get network IDs
 
That wouldn't help - SE thinks the accounts (which are unregistered) are separate: stackoverflow.com/users/7639466/gamesolver and gamedev.stackexchange.com/users/98362/gamesolver
so right now they have different network IDs.
 
@Glorfindel I know, but in other situations it would
 
Could we tweak the thanks/praise filter to catch this answer?
 
@Glorfindel could you make an issue on GH? I'll get to it when I can
 
9:36 AM
@angussidney I'll try it myself then, it doesn't look too difficult once you understand the regexes.
 
New metasmoke user 'mrjink' created
fp- by Cerbrus
 
Genuine question containing "muscle transformation"... I never thought I'd see the day
 
[ 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: /julo-eye-cream/ by edithfisher on workplace.SE
tpu- by FelixSFD
[ SmokeDetector | MS ] Bad keyword in body, 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: AM Cream Revitalizing Skin Therapy by user227117 on askubuntu.com (@ThomasWard)
tpu- by Glorfindel
 
@SmokeDetector k
 
Again an AskUbuntu post which didn't grey out ... but I was stupid enough to reload the page.
 
9:44 AM
[ SmokeDetector | MS ] Bad keyword in title, blacklisted website in body, link at end of body, pattern-matching product name in body, pattern-matching website in body: making you feel muscle x tst 1700 by lliottarah on drupal.SE
tpu- by FelixSFD
 
@SmokeDetector k
 
[ SmokeDetector | MS ] Bad keyword in body, pattern-matching product name in body, pattern-matching website in body, repeated URL at end of long post: Creatine Despite the fact that there are presently by Thommeh on apple.SE
tpu- by FelixSFD
[ SmokeDetector | MS ] Few unique characters in answer: FPDI free parser that supports PDF version higher than 1.4 by Muruga on stackoverflow.com (@dorukayhan)
[ SmokeDetector | MS ] URL in title, bad keyword in body, link following arrow in body, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: platinumcleanserinfo.com/cleara-bella/ by lerykeaneth on apple.SE
tpu- by Floern
[ SmokeDetector | MS ] Bad keyword in link text in body: Read concern level of majority error while upgrading Sharded Cluster from 3.2 to 3.4 by umarxe on stackoverflow.com (@dorukayhan)
fp- by tripleee
 
not allowing "replica" is a bit too harsh, isn't it?
 
Seems fair enough imo
could be better but fair
 
actually many of those don't actually trigger this filter
 
9:56 AM
I think it has a lot of legitimate uses as well ... those posts aren't even in MS.
 
[ SmokeDetector | MS ] Link following arrow in body: Multi select is not working, and gives error 'this.validator is not a function' in angularjs2 by test on stackoverflow.com (@dorukayhan)
fp- by Gothdo
 
10:18 AM
[ SmokeDetector | MS ] Bad keyword in body, link at end of body, pattern-matching website in body, repeated URL at end of long post: Lovely La Femme Dresses Match Lovely by Kyszsloi on askubuntu.com (@ThomasWard)
tpu- by Gothdo
[ SmokeDetector | MS ] URL in title, bad keyword in title, blacklisted website in body, blacklisted website in title, pattern-matching product name in body, repeated URL at end of long post: offersreviews.org/max-muscle-t1000/ by foay lola on apple.SE
tpu- by FelixSFD
[ SmokeDetector | MS ] Pattern-matching website in body, repeated URL at end of long post: Me just for a while just by nifewedayo on graphicdesign.SE
 
@SmokeDetector k
 
10:41 AM
[ SmokeDetector | MS ] Bad keyword in link text in body: Stuck on splash screen error loading shared library by P.B. on askubuntu.com (@ThomasWard)
fp- by Gothdo
 
10:56 AM
TIL cdn.rawgit.com has quite aggressive caching
 
[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body: Benefits Of Herbal Enlargement Products by sreciolia on apple.SE
tpu- by Gothdo
 
@SmokeDetector k
Now that I have 166 accounts, I have to get 200 rep on all of them...
 
[ SmokeDetector | MS ] Blacklisted website in body, link at end of body, repeated URL at end of long post: Secret allure cream Pros: secret allure cream by yojijiyoz on superuser.com
tpu- by Floern
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, link at end of body, repeated URL at end of long post: Are Beard Czar Negative effects Dose? by czarbeard1 on superuser.com
tpu- by Mithrandir
[ SmokeDetector | MS ] URL in title, bad keyword in body, bad keyword in title, pattern-matching website in body, pattern-matching website in title: powerupmale.com/intellex-plus/ by PaulButlerr on graphicdesign.SE
tpu- by Gothdo
[ SmokeDetector | MS ] Blacklisted website in answer: Initializing Games Client in Android by Govind Dhera on stackoverflow.com (@dorukayhan)
tpu- by Gothdo
 
J F
11:16 AM
@Cerbrus FYI it’s [ci skip], not [ci ignore].
 
Yea, derped once or twice
 
[ SmokeDetector | MS ] Blacklisted website in body, repeated URL at end of long post: used to look more bursting than my age by cabiha24 on graphicdesign.SE
tpu- by Gothdo
 
@JF: data.is_answer = data.link.indexOf("/a/") >= 0; --> linter says it should be camel case... But that'd be the only camel case property
Is there a "ignore this line" comment?
 
[ SmokeDetector | MS ] Link at end of body, link following arrow in body, pattern-matching website in body: Ultimate Testo Explosion - Improve your Blood Ciculation by ronalakreddy on askubuntu.com (@ThomasWard)
tpu- by Glorfindel
 
J F
@Cerbrus // eslint-disable-line camelcase
 
11:20 AM
Thanks :D
 
!!/alive?
 
@AshishAhuja Of course
 
there's a wut command? I was just going through chatcommands.pys source..
!!/wut
 
@AshishAhuja Whaddya mean, 'wut'? Humans...
 
11:33 AM
!!/alive??
 
@AshishAhuja Of course
 
@Glorfindel not sure
!!/alive?!?.?
 
@AshishAhuja Of course
 
!!/ali?ve
ok, I understood
!!/?alive
 
[ SmokeDetector ] SmokeDetector started at rev 13aae02 (angussidney: Strip code blocks --autopull) (running on teward/Aurora)
Restart: API quota is 4884.
[ SmokeDetector | MS ] Bad keyword in answer, bad keyword with email in answer, email in answer: What are some of the immediate challenges to break through before finding a cure for mad cow disease? by Rose Weel on biology.SE
tpu- by Gothdo
[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body, pattern-matching product name in body, repeated URL at end of long post: effective plumping outcomes on the pores? by mikenjibond on stackoverflow.com (@dorukayhan)
tpu- by tripleee
 
J F
11:43 AM
@Cerbrus If you call the GitHub API, you can get the latest commit and fetch that from cdn.rawgit.com, like I do here.
 
[ SmokeDetector | MS ] URL-only title, bad keyword in body, blacklisted website in body, blacklisted website in title, link at end of body, repeated URL at end of long post: menintalk.com/priamax/ by yenyperelta on askubuntu.com (@ThomasWard)
tpu- by Gothdo
 
@JF Sweet, that works
 
[ SmokeDetector | MS ] Bad keyword in body, blacklisted username, blacklisted website in body, link at end of body, pattern-matching product name in body, repeated URL at end of long post: radiant and flawless skin. This will help consequently improve on your self by ksaryjuriya on security.SE
tpu- by Glorfindel
[ SmokeDetector | MS ] Bad keyword in body, link at end of body, pattern-matching product name in body, pattern-matching website in body: Where to Buy Ultra Slim 400? by Ronaldmugal on workplace.SE
tpu- by Gothdo
 
11:59 AM
^ Why this was autoflagged with only one flag?
 
Workplace
That site requires just 3 red flags to nuke a post, so we do only 1.
Same goes for EL&U.
 
[ SmokeDetector | MS ] Bad keyword in body, pattern-matching product name in body, pattern-matching website in body, repeated URL at end of long post: Incorporate boosting vitality generation, enhancing by lindarobbins1212 on graphicdesign.SE
tpu- by Glorfindel
 
@Glorfindel Have that been changed recently, or have it always been like that?
 
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, blacklisted website in body, link at end of body: Concerning of Beard Czar & Components! by czarbeard1 on workplace.SE
tpu- by Gothdo
[ SmokeDetector | MS ] Username similar to website in answer: Installed laptop-mode-tool 16.04, now USB mouse not working after rebooting until it is re plughed in by RogTek on askubuntu.com (@ThomasWard)
fp- by Gothdo
 
12:08 PM
@SmokeDetector put in suggested edit
 
12:32 PM
apparently FDSC report isn't working either..
 
[ SmokeDetector ] SmokeDetector started at rev 13aae02 (angussidney: Strip code blocks --autopull) (running on teward/Aurora)
Restart: API quota is 4522.
[ SmokeDetector | MS ] Non-Latin link in answer: Usar Ajax.BeginForm MVC 4 aspx c# by Islame Mando on es.stackoverflow.com
tpu- by Glorfindel
 
@SmokeDetector That threw me off, flag dialog in a different language
^ I keep thinking @Undo is messing with my script.
 
CI on 4b807ba succeeded.
 
@Cerbrus I've learned to read it in Japanese and Russian
I guess Korean will be coming up next
 
I can read it in Russian, Spanish and Portuguese; for Japanese and Korean I rely on muscle memory.
 
@SmokeDetector offensive, probably a troll?
 
Member for 6 months on SO.
 
flagged VLQ
 
So a very patient troll, then.
My edit was accepted (by the OP) @tripleee @Gothdo @Cerbrus you might or might not want to retract your offensive flag here.
 
hmmm, still suspicious, there never was Linux 0.0.1 and the kernel is an awful place to start understanding basic CPU architecture
 
12:54 PM
[ SmokeDetector ] URL in title, bad keyword in body, bad keyword in title, pattern-matching product name in body, pattern-matching product name in title, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: www.nutritionfit.org/max-grow-xtreme/ by naluvaj24 on graphicdesign.SE
tpu- by Glorfindel
 
looks like there is a thing called Linux 0.0.1
 
@Glorfindel trying to convince some SE dev to randomize the order of flags in non-english flagging menus
 
someone ping me if Matthew Read turns up
@MatthewRead chat might be a better venue for this discussion; feel free to drop into Charcoal HQ to talk about it. I'd like to understand why you think this is abuse. — ArtOfCode 3 mins ago
 
@ByteCommander then we'll make a script to sort them again, and whine on meta
 
12:57 PM
@ArtOfCode So far the only outright negative response, no?
At least on the thread
 
@Magisch pretty much, yeah
surprising, really
 
if username=="Glorfindel":
    flags = shuffled(flags)
 
@ArtOfCode In over 12k views
 
I'm all for having separate accounts just for the flagging, actually
 
We've been featured for almost a week now
 
12:58 PM
@ByteCommander Это не поможет.
 
Sorry, I don't buy anything.
 
@ByteCommander E: flags is immutable
 
now it doesn't matter
 
naa- by ByteCommander
[ SmokeDetector | MS ] Bad keyword in body, link at end of body, pattern-matching product name in body, pattern-matching website in body, repeated URL at end of long post: In a matter of seconds, by neginimiya on superuser.com
tpu- by Gothdo
 
@SmokeDetector In a matter of seconds this spam question is gone.
Forgive me the use of non-idiomatic English.
 
1:09 PM
[ SmokeDetector | MS ] Few unique characters in answer: Laravel project missing .env file by this is bogys on stackoverflow.com (@dorukayhan)
[ SmokeDetector | MS ] No whitespace in answer: Calculate count records from mysql group by fields by fghgfhgf on stackoverflow.com (@dorukayhan)
tpu- by Glorfindel on Laravel project missing .env file [MS]
tpu- by Glorfindel
[ SmokeDetector | MS ] Bad keyword in body, link at end of body, pattern-matching product name in body, pattern-matching website in body, repeated URL at end of long post, blacklisted user: The different segments by neginimiya on superuser.com
tpu- by Glorfindel
 
1:27 PM
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, link at end of body, pattern-matching product name in body, pattern-matching product name in title, pattern-matching website in body: where to buy max grow xtreme by taraguire on drupal.SE
tpu- by Glorfindel
[ SmokeDetector | MS ] Bad keyword with a link in answer: How to show the bow string stretching animation (?) by Henrietta Weast on gamedev.SE
tpu- by Glorfindel
New metasmoke user 'Daniel Davies' created
[ SmokeDetector | MS ] Non-English link in answer, non-Latin link in answer: Spelling of feminine name "Майя" in Ukrainian by NogrTL on ukrainian.SE
 
@SmokeDetector I'm sure I disabled those rules.
!!/rev
 
!!/pull
 
[ SmokeDetector ] SmokeDetector started at rev 4b807ba (ArtOfCode: Merge pull request #563 from Glorfindel83/master) (running on teward/Aurora)
Restart: API quota is 4109.
 
@ArtOfCode Opens can of coke for Art
 
1:41 PM
Is this regular coke, or christmas coke (same liquid, different labeling)?
 
Most of the time, !!/pull opens a can of worms instead.
 
New metasmoke user 'Josef Van Zyl' created
 
@Magisch At the moment, the filters are nowhere near state of the art. It's mostly a bunch of regexes from what I see. This is not scalable and can't be but a temporary solution. — Kuba Ober 1 min ago
Maybe you're more qualified to talk about that art
 
Smokey may not be "state of the art", but it is "state of the Art". That's enough for me.
10
 
not sure what he's referring to either
what else but patterns can we use?
Arcane machine learning? We already know thats far off for now
 
1:50 PM
Maybe he wants to see fancy neural networks and stuff
 
Those would be cool
But wait, we do have a bunch of neural networks already. They monitor Smokey's progress relentlessly.
 
Meh, I somehow prefer predictable regexes over unpredictable neurons.
 
One thing I guess is that he is referring to the fact there are many different types of spam, and with patterns you could never cover them all... but thanks to SO restrictions on new users, the spam we get tends to follow a (relatively) small number of patterns so it's not a problem here
 
... and predictable regexes have been doing us unpredictable neurons a great service up until now and on
 
[ SmokeDetector | MS ] Bad keyword in answer: Question about Sicario interrogation scene by Mike Perry on movies.SE
tpu- by Glorfindel
[ SmokeDetector | MS ] Link at end of answer: How to create a responsive hamburger menu in AMP HTML by Bughinese on stackoverflow.com (@dorukayhan)
naa- by Floern
New metasmoke user 'Empty2k12' created
[ SmokeDetector | MS ] Bad keyword in body, bad keyword in title, pattern-matching website in body: What is Andronite Enhanced? by maysslak on drupal.SE
tpu- by Glorfindel
 
2:13 PM
in SO Close Vote Reviewers on Stack Overflow Chat, 1 min ago, by Petter Friberg
Yeah killed before 7 flags (only 5), wondering when they will fix smokey...
 
[ SmokeDetector | MS ] Bad keyword in body, blacklisted website in body, pattern-matching website in body: Metaxlim Reviews by alissono1588 on apple.SE
tpu- by Glorfindel
[ SmokeDetector | MS ] Link at end of answer: Using remote host's USB port as local USB (Linux and Win) by CarlosRPEvertsz on stackoverflow.com (@dorukayhan)
naa- by Gothdo
 
2:48 PM
[ SmokeDetector | MS ] Link at end of answer, non-English link in answer, non-Latin link in answer: Does while(1){} do the same as while(1); by ashkan on stackoverflow.com (@dorukayhan)
 
@SmokeDetector k
 
[ SmokeDetector | MS ] Offensive answer detected: How to get a list of all extended properties for all objects by john on stackoverflow.com (@dorukayhan)
tpu- by Magisch
 
3:07 PM
[ SmokeDetector | MS ] Link at end of answer: Get ID3 tags from stream on iPhone by annsmith on stackoverflow.com (@dorukayhan)
New metasmoke user 'Kirk' created
tpu- by DavidPostill
 
I prefer Picard
 
@SmokeDetector f
 
Guys, I've got a bit of a draft for a userscript that would allow feedback / flagging from chat, with an "nice" GUI / popup, so you wont have to go to a new tab to see the report. Of course, none of the actual feedback logic is implemented, but I'd love some feedback. It adds a button (🛠️) to the info AIM adds (It requires AIM), the rest should be pretty self-explanatory. ([1]-[3] selects feedback options [Esc] closes it)
 
@Cerbrus warning: StackExchange doesn't like being inside an iFrame
 
3:20 PM
Not an iFrame
The report also contains the reported posts' HTML
Which I dump into a div, to build the popup.
Oh, and all links in there show their url on hover (replacing the text)
 
oh so we now have faded posts that are deleted?
 
3:36 PM
@Cerbrus as a POC, it's good. Things: (a) Make clicking outside the popup close it. (b) Showing link URLs on hover is a nice idea, but breaks when the URL is shorter than the link text (flickers between the two as the link moves in and out of focus). I'd make it a tooltip instead. (c) FR: Show the site's header image above the post title/text. (d) Show if it's a question or answer.
 
@YvetteColomb yes, except for AskUbuntu posts #knownbug
 
thanks
 
(also, questions shouldn't have the NAA button available)
 
@Cerbrus fp should be fp-, and we need tp- for vandalism
 
[ SmokeDetector | MS ] Username similar to website in answer: Adding a privacy policy in application in the Google Play Market by TermsFeed on stackoverflow.com (@dorukayhan)
tpu- by Gothdo
 
@ByteCommander I've actually trained two single-unit GRUs and so far they're both doing pretty poorly (although it seems to be generalizing well at least):
TPs: 28341/46142 FPs: 6208/9273 NAAs: 683/2317 Total: 35232/57732
TPs: 29738/46142 FPs: 7228/9273 NAAs: 0/2317 Total: 36966/57732
The first one was trained without batches, the second one was trained by making mini-batches based on sentence length. (I have no idea why it isn't identifying any NAAs.)
 
@YvetteColomb An answer to the question?
 
I might try a stacked LSTM but they're such a pain to train.
 
the links. The images with links. so big and all an off topic question to boot
 
(cc @Magisch on arcane ML)
 
3:49 PM
[ SmokeDetector | MS ] Bad keyword in answer: Should a deployment user have his shell set to /bin/false? by jons usa on serverfault.com
tpu- by Gothdo
 
can people not flag for a minute, to give me time to enter my code sheesh
 
@ArtOfCode Have to say, this has been one of the more surprising meta posts in terms of people's reactions.
 
@Undo aye, I was expecting at least a few negative reactions
 
What have others tried in terms of ML? I hadn't heard about that
 
@quartata A few things, but nothing good enough to share
 
4:01 PM
@quartata I did a Bayesian thing a while back; look for the sdml repo. It was terrible, before you ask. I think @tripleee was looking into doing something at one point.
 
I can get metasmoke data for anyone playing with that, though
 
I already got all the data I needed, look up :P
I might cut the NAA class out
That would open it up to other architectures
(namely SVMs)
 
@ArtOfCode Thanks, I'll add them to my todo list :D B/D have been on my mind already :-)
@DavidPostill thanks, fixing
 
I think the strangest part is that it's most accurate on false positives....
I suspect I've weighted the loss function weird.
 
New metasmoke user 'user2428118' created
[ SmokeDetector | MS ] URL in title, pattern-matching website in body, pattern-matching website in title, repeated URL at end of long post: onlinehealthmarkets.com/secret-allure-cream/ by secret allure5 on apple.SE
tpu- by Gothdo
 
4:10 PM
Hola
review/feedback seems broken, I'm getting a 500 error
AbstractController::DoubleRenderError in ReviewController#add_feedback

Render and/or redirect were called multiple times in this action. Please note that you may only call render OR redirect, and at most once per action. Also note that neither redirect nor render terminate execution of the action, so if you want to exit an action after redirecting, you need to do something like "redirect_to(...) and return".
Extracted source (around line #35):

#33 # Check for double render errors and set the content_type after rendering.
 
@ArtOfCode yeah, I have a fork with a multinomial classifier but I haven't revisited it recently
 
Damn, "🛠️" doesn't render for me on my home PC :-/
 
I was collecting input data for the classifier but it takes some manual work to put the labels I want
@Cerbrus I see an empty box, which character are you trying to use?
 
"Hammer and Wrench" emoji
Apparently windows 7 doesn't support that character set
 
nvm, just saw the starred msg
 
4:17 PM
it's not about the character set, it's about having a font which includes the glyph
 
Hmm
 
the character set is simply Unicode
fileformat.info/info/unicode/char/1f6e0/index.htm says introduced June 2014 so it's not like it's spanking new
 
@Cerbrus Works for me on Windows 7/Firefox
 
Weird, maybe I need to update something here
 
@vzwick That'd be because you don't have reviewer privileges and we don't handle it correctly.
Fixed now, thanks :)
 
4:28 PM
@quartata my thinking is to try to identify particular categories of spam, rather than just lump everything into the current categories TP, FP, and NAA
the "replica" hits we looked at earlier today would seem like a good category which we are currently not identifying successfully
the pharma spam is another set, although we already have a very high detection rate for that
but it would reduce the need to manually add conspicious keywords to the filter
if not completely eliminate it
but in order to pull that off I need properly labelled sets of these
as well as a sizable corpus of non-spam posts just to avoid overfitting
the latter I actually have, but it's only recent posts from whenever I happened to run the collector
but probably good enough for jazz
now I just need to find the time to sit down and train it and tweak it
 
@tripleee SEDE
 
@Undo using the Metasmoke API to fetch recent posts in exactly the format I need is better IMHO, I don't need to second-guess how various HTML things are encoded in the JSON etc
 
Or download the data dump and scrape the top 10k off each site
 
I have something like 10k total, I think that's probably enough for the main sites, but sites which get very little traffic are obviously not well represented
but like I said, probably enough at least for a prototype, if I can find the time
 
@tripleee SEDE's Posts.Body should be the same exact format we store in metasmoke, fwiw
Yeah, even if you just limit it to the top n sites we see spam on at first it'd work
 
4:55 PM
[ SmokeDetector | MS ] Bad keyword in link text in body: WOWZA LiveAutoRecord by Atul on stackoverflow.com (@dorukayhan)
 
@SmokeDetector why
 
@Gothdo Post - Bad keyword live-stream in link text
 
FDSC is broken.
 
[ SmokeDetector | MS ] Few unique characters in body, title has only one unique char: --------------- by rem_LapLB on stackoverflow.com (@dorukayhan)
 
Doesn't seem to be sending feedback to MS
or at least it's not showing up here
 
4:57 PM
@SmokeDetector v
 
And MS API isn't sending the body again
@ArtOfCode everything is wrong
 
Did your filter change?
 
New metasmoke user 'Joao Duarte' created
 
No. Art confirmed that that was a bug yesterday but it had seemingly gone away
 
Have I got another browser color bug, or did GH make a bunch of colors more vibrant
 
5:08 PM
New metasmoke user 'Edvin Tenovimas' created
 
@Magisch sssh don't shout about it
I'm just gonna let it run and see when the CMs decide to unfeature it
 
lol
 
if they wanna leave it there for two weeks, mkay, more people for us :)
 
viewed

12583 times
@ArtOfCode I just now realize you failed to make the best name pun on the state of the art comment
D:
 
[ SmokeDetector | MS ] Username similar to website in answer: Grohe Ladylux kitchen faucet leaking underneath by GROHE on diy.SE
naa- by Floern
 
5:25 PM
sd n
 
@Undo [:35765292] That message is not a report.
 
maaaybe those messages shouldn't count
@ArtOfCode Noticed that too this morning. Either we're both gaining eyesight, or they did something
 
[ SmokeDetector | MS ] Pattern-matching website in body: Secret Allure Cream by bvsv234454 on stackoverflow.com (@dorukayhan)
tpu- by Floern
 
@Undo that has been coined before. The problem is that people who know to do sd - n in these cases need to adapt as well, or risk the chance of feedback-ing the wrong report.
 
yeah
 
5:32 PM
[ SmokeDetector | MS ] Bad keyword in link text in answer: Live stream recording in Wowza Streaming Engine by Tim Dougherty on stackoverflow.com (@dorukayhan)
 
5:56 PM
@Undo if it's changed, it's probably best to keep the current functionality and use sdn for a new feedback scheme.
 

« first day (1200 days earlier)      last day (2619 days later) »