[ SmokeDetector | MS ] Bad keyword in title, bad keyword with email in body, email in body: we offer loan at low rate any one interested should contact us by Victor on money.stackexchange.com

@SmokeDetector k

sd f

!!/alive

@Ferrybig Of course

Well...work day is done. The Oracle database just died. Thankfully, I'm not support. Someone is going to have a bad afternoon. ponders what to do with all this new found free time

@Andy perks of working in a big corp

If that happened for us it would be a "nobody even remotely related to IT leaves this building until it runs again" type thing

[ SmokeDetector | MS ] Few unique characters in answer: Android - Get index first and last same item in ArrayList? by YaMiN on stackoverflow.com

@SmokeDetector why

@HenryWHHackv2.0 Post - Contains 4 unique characters

@HenryWHHackv2.0 It's answer, not question. It was self-deleted by owner.

What if someone posts a spam/rude/abusive post, then immediately deletes its, and then after an hour undeletes it? Would SmokeDetector catch it?

@Gothdo Wait are you talking to me?

@Gothdo Not currently

[ SmokeDetector | MS ] Blacklisted user: Toyota Corolla XLI ABS Issue by Jani Jaaz on mechanics.stackexchange.com

@SmokeDetector k

@HenryWHHackv2.0 No, that question wasn't for you.

!!/report money.stackexchange.com/a/69596 money.stackexchange.com/a/69597 money.stackexchange.com/a/69599

[ SmokeDetector | MS ] Manually reported answer (batch report: post 1 out of 3): Get tax deduction for expensive car expense by Marin on money.stackexchange.com

[ SmokeDetector | MS ] Manually reported answer (batch report: post 2 out of 3): Home office deduction using simplified method & expensing of non-permanent office modification? by Marin on money.stackexchange.com

[ SmokeDetector | MS ] Manually reported answer (batch report: post 3 out of 3): Is there a tax deduction for renting office space in service of employer? by Marin on money.stackexchange.com

@SmokeDetector k

@SmokeDetector k

@Gothdo The first one has the same link in the source, btw

Actually, there is a 4th post which seems ok to me.

Also, why does Smokey order reports differently from what I had in the command?

@undo You're aware FDSC doesnt work atm?

That being?

The userscript.

@Andy they probably tried to upgrade to 16.04

@DEAD Art's

I assume you're logged in?

Well, I wasn't the one pinging you.

@undo I don't have a tab open

Shouldn't matter, as long as there are valid cookies

@undo Well, right now my only pet peeve is how the userscript gives an error every time I flag something as NAA.

I'm going down the highway trying to debug this on my phone. Should be able to actually see what's happening in a few minutes

Don't debug and drive.

Debugging while driving kills

@undo Heh. During last Thursday's conversation with Pops I was traveling 80mph down the interstate.

(not driving)

Both of y'all better be/have been in not-the-driver's seat.

Flying?

Passenger seat

@DEAD Have you paid for internet access on a plane? I'm already broke because I bought the ticket.

Well, last I flew with a plane, I wasn't able to spell internet correctly.

Speaking of planes, American had Deadpool on their inflight, free wifi, a month or two ago. I decided to watch it...then learned that it was "edited for content". The editing, though, wasn't to remove objectionable scenes. It was for time limits and who ever did the editing, wasn't good at it. They just chopped scenes off mid-sentence.

I don't think it's possible here, even

@Andy That's worse than not watching it, argh

I just nominated in ELL's mod election.

@undo Your server downtime logged me out :o

:o

Since you popped in, @Pops, have you had a chance to find out whether or not you can share more details on how spam detection works on your side?

lurking CM is best CM

@Andy Yes, sorry, that's actually what I came in for, but then I got momentarily distracted right afterwards.

That information would be pretty neat as then we could start filling more of the holes still in that system

It's because you were concerned for our well being, wasn't it? I appreciate the concern ;)

@Magisch No one is as good as Shog in that regard.

@DEAD WHAT YOU SAY? Bee arr bee, opening 200 tabs...

(Still typing about the spam stuff, just a few more minutes.)

Try harder

@Pops Point being, Shog is always watching. If you mention him, most of the times he'll magicly appear and then ethereally disappear again

Shog has 1000 tabs open prolly.

In five different browsers.

Restart: API quota is 4081.

!!/errorlogs 20

Unfortunately it's not much. A big part of what we're doing involves looking at what we're seeing/have seen in the past, and making notes (i.e. incrementing a bunch of numbers in key-value stores) about stuff like, for example, IPs. We then use those notes to decide how to behave going forward.

One specific problem we're looking at right now has to do with spam coming in from many different places at once (AKA snowshoe attacks). So if we get (made-up numbers incoming) 1000 attempted spam posts in a day and catch 99%, the other ten posts are still making it seem to the wider community like we're not doing anything to stop spam, especially if they linger on the front page.

So I liked that idea from the other day about getting rid of spam faster once it's appeared, but I'm also interested in ways to catch it between when it's submitted to us and when it shows up on the sites (i.e. preventing it from appearing to the vast majority of users). As far as Smokey is concerned, my gut reaction is that your "this is probable spam" heuristics (regexes, I believe?) may be helpful there, but I haven't given that serious thought yet.

@Pops We do it roughly the same way

We look at common patterns in confirmed spam posts, and make filters based on that

What about making suspicious posts not show up on the main page for 10 minutes (and make it effective link-only), and feed it in a special websocket so smokey could scan it with the team, before it appears in the main site

One such heuristic is a website blacklist, another one is too few unique characters. Another one are common spam words like garcinia or health review, in combinations

there are hundreds

@Pops If you trust us to not misuse that, you could make a provision to let posts from 1 rep users (>95% of spam comes from them) doesn't show on the main page until we had a chance to scan them, and then allow these posts to be spam flagged and disposed of before they ever make it to main

Do you treat all sites the same, or are there exceptions for certain sites (the example of "garcinia" made me think maybe Skeptics or Health could have legit questions including that term)?

we have a few exceptions IIRC

There are rules that are ignored per site

but most of the false positives are handled via feedback

@Pops we cast a wide net (purposefully) because getting wrongly caught by smokey means nothing besides 3-5s work for one of us reviewers

for instance, we catch links at the end of posts.

Its all more or less based on the common sense of the developers, and that seems to be strangely effective. Websites and user name patterns that spam often (like the wingding guy or the %juriyam% guy) are blacklisted, and we catch tons of posts that way

the main strenght of this is that if a specific guy started spamming his BS health product today, we could catch every instance of it going forward as soon as one of the dev's gets a look at them, and before based on user accounts.

@Magisch that'd be overkill.

(afk, but wanted to get that out)

@undo maybe, just tossing ideas atm

@Magisch well... I logged everone out, on purpose.

@Pops Possible, but that's (1) dev work, and (2) a few minutes is a long time, really.

Also, we'd have to be careful not to make the whole network rely on... ya know, a raspberry pi in my closet being alive.

This is an odd situation that I don't have an answer for yet. On the one hand I agree with what someone said last week that there should be human eyes on this process; we're not advanced enough yet to let spam detection be 100% automated. On the other hand, I'm kinda hesitant to build any system that relies on any small group of humans to vet all posts.

This is absolutely not about you; the work y'all have done on Smokey is incredible--nearly literally so--but over years, people's life situations change, and it would be... awkward, at best... to find ourselves in a spot where you current people have all moved on (next level of school, harder job, more kids) and no new reviewers have stepped in to take your place.

@Pops, Of that spam that makes it to the site and then is removed, what's the break down of rep for the posting users? My guess is that it's mostly 1 rep, but I have no data to back that up.

Also that thing about the status of undo's closet.

@Pops It could be on a as-long-as-it-works basis

@Magisch That's not really what you build a production system on though for the long term.

@Andy That's a good question and I don't have an answer for you offhand. I can't remember ever personally seeing spam from a non-1-rep account.

And it could be little things like giving smokey a higher api quota so we don't have to bundle requests ... or make confirmed SD reports take less flags

The less flags thing in particular would make it that throughout the work day, much of the spam could be killed off within the first 5 views

@Pops I've seen some ... its almost always cases that require manual intervention though. We're very hesitant to nuke (redflag) users that have any legit posts whatsoever

@Pops This is very real. I'm not terribly concerned about it (I've seen the turnover, and it's fairly positive), but it's real

Yeah, the less flags idea was what I was referring to as the "idea from last week"; it's sort of "big site thinking" though, even five views on some of the really quiet sites can take a while, especially because users there often aren't familiar with how flagging works (much less how spam flags are different).

I can understand you guys not wanting to invest many ressources into a solution that may not be permanent @Pops, but most of the improvement ideas require very little involvement from SE

@Pops Often there is 3+ users in here flagging. Less flags (say, 3 for confirmed reports) would mean a lot of posts on smaller sites would get nuked before anyone coming through on the site sees them

@Pops Assuming it is almost exclusively 1 rep then, is the possibility of introducing a short delay for new users' posts appearing on the front page an option? If a post still appears on the websocket smokey can look at it, do it's thing, and if it doesn't get spam flags in X minutes, it appears on the front page as normal.

we see nearly every report

@Pops Ya know, one possible idea if we're that concerned about timing is to let moderators across the network nuke spam.

^

@Andy Yeah, I did like that idea, gonna have to figure out a way to see how effective it would be.

On SO, spam flags rarely go thirty seconds without someone seeing it.

@undo Heh that was exactly the thought I had for pairing with the "likely spam, delay posting" idea.

@Pops I see more rude/abusive than spam from more than 1 rep accounts, mainly rages, usually comes from the higher reputation people

Its almost unfair to spammers on SO sometimes

@Pops We do have some stats on how quickly posts are removed once reported. @ArtOfCode, is that information available via MS's API?

I know there are concerns about the principle of letting moderator privileges 'leak', but I don't know a better way to get trusted people to look at stuff.

But then again it's a small group of people to lean on.

Like they just posted their elaborate spam they spent time snowshoing around spamram for ... to get nuked in less then 20 seconds

@Pops small, but tied to the site itself instead of a little auxiliary group like we have.

If the SO team falls off a cliff, we've got bigger problems than spam deletion timing.

I'm not a moderator, and I'd like to think I'm fairly competent in reviewing posts by now

@undo That was a strange way to phrase that. Are you aware of an imminent cliff-related threat? Blink twice (in, uh, the TL, I guess) if you need help from the Aviation mods!

I plan on doing this at least 3-5 hours a day for at least another 2 1/2 years for certain

@Magisch Sure, but as far as the system is concerned you're on the same level as a... whole bunch of other people. We've already got a giant vetting process for moderators. It keeps some great people out, but it also keeps a lot of not good people out.

I know

I've got over 6k completed reviews in the queues, I've cast thousands of close votes, over 2700 votes total, I have over 1,5k helpful flags across the network. But jonny mcFGITW with his 10k points is more trusted then me by the system

Ah well, every system has outliers

Anyway, it's something to think about @Pops. I strongly doubt that adding network spam flags to SO's queue would double it, and it's already not bad at all.

@Magisch This.

So much that.

If you added SO posts to other site's queues it might give people heart attacks, but you don't have to do that.

sd why

eh?

sd f

@undo [:31730522] That message is not a report.

@undo metasmoke.erwaysoftware.com/post/37407

@SmokeDetector f

!!/rmblu money.stackexchange.com/u/47481

@undo User is not blacklisted.

I see no evidence of spam there.

It has a spam link in it

The "write off your car" link was spammy

@Magisch Eh, don't feel bad. These are not hipsters producing artisinal handicrafts who are struggling to reach a customer base that just doesn't know they're there. (Although, the reality is probably also sad, for different reasons.) Here, have a mayonnaise distraction.

oh, hold on.

Did someone edit that out?

@Pops As an older person with kids the same age and older than many of the regulars in here I understand what you're saying about life circumstance changing and hence losing people from the pool. However, there have been people active on the site for long enough and a pool big enough to replenish as people do move on. With SE employees overseeing it, you would have ample warning if there was a diminishing population in here.

...wow, the room really moved on without me while I was looking up the link for that mayo store.

@Pops We go from banter to business fast when there is potential spam to be nuked

Otherwise SE can pay employees to do it ;)

@Pops So anyway, sounds like we have some things to think about.

@GregP Yes, it's probably safe, but if we can build a more robust system at roughly equal effort, I'd rather do that, is really all I'm saying there.

For many of the proposed solutions the effort on SE's side (besides deciding that it should be done) is nearly zero

While you're around... is there any way I can poke meta.stackexchange.com/q/279993/215468?

Hmm, that title looks vaguely familiar. Hang on.

@GregP It'd be possible to build something that doesn't rely on humans.

I don't see why we couldn't phase the humans out, at least for the purposes of this. We're hitting 99.x% on some reasons.

@undo GitHub is attempting to do so

@undo Any system will need a review to determine the accuracy, unless you go by the community flags.

@GregP Community flags and moderator response. We just need to get that data.

there are better ways to do it.

remember, feedback started as a hack. It is a hack.

(bad network)

@undo but also the changing face of spam and keeping up with that. The system will always need tweaking

And we can do that (barring ML stuff, which is becoming very real).

What we don't need is a dedicated group of people in here sending feedback.

There is no such thing as perfection when you're trying to keep up with human deviousness and duplicity

Of course not. No such thing as perfection with human reviewers either.

There will always be the need for some kind of review taskforce. I believe. This can be done on a site wide basis by introducing a way of reporting and reviewing spam. An extension of what we're doing here, but for all users to have access (with some rep restrictions)

or with a paid SE people

Hooold on.

or some group that are willing like us, or much like socvr targets close votes

Why can't we just use the data SE already has, again?

that's useful for a retrospective look

If something is deleted as spam, it's a tp. If it's declined, it's an fp. Easy.

We can do that in realtime.

going forward none of know what we will need to be looking at

yes I agree

huh?

Having a hard time parsing that sentence

but there will also be things that are missed

Well

And that's what false negatives are for. We can do that too, with a little work.

currently it works as it is always being tweaked

@undo Our biggest stranght is that we can adapt to changes in spam patterns within minutes or hours

Yes, of course we need folks to take care of it.

it's being tweaked by human beings watching the site

that is what I'm saying

We don't need folks sitting in here flagging and sending feedback, though. Ideally that wouldn't even happen

but it can be done to be site inclusive, rather than small group inclusive

Ideally we wouldn't even post to chat. Run everything directly into the flag queue.

If there was an obvious way to fully automate this, SE would have done it already.

the flag queue is manned by moderators

yes of course, it would need to be more refined that the current set up

How many mods do you want?

I am saying that se is going to need to keep some taskforce on it, to keep it successful

@Magisch mods for what?

Sure. We don't need flaggers or people sending feedback, though

Your idea would mean that all of us non mods don't get to participate anymore in any meaningful capacity. Not sure if thats what you want, but thats what cutting out chat and giving it to a flag queue would do.

Just send it straight to the flag queue, record the result.

And then? You're in machine learning territory again

@undo I agree with that

@Magisch If we can work ourselves out of a job, we win.

Unless you want mods handling the flags to come up with the pattern

But then there is the issue of catching stuff that is being missed, or the time it sits before being caught

@GregP And that's the problem you solve by showing spam flags outside of their originating site. That's a network problem, not our problem.

really there is still the need for a chat room and pings, so people know to flag

Idk, I like cleaning up the site. I would be a bit sad if that work got offloaded to mods and I would be unable to help

@Magisch you'd find a way ;)

I happen to know a place with a bunch of plagiarism to find

@undo well notifications in chat are a good way to do that, as it will find anyone currently in the room

Okay, here's something that is disturbing me: We can't get wrapped around the current way of doing things.

I'm just thinking, yes the accuracy can be given by whether the post is deleted or not. So that would be the f or k

I'm getting a strong vibe of self-preservation. I'm not going to let that slow us down.

All this stuff here, SOCVR, this room, all this, its so nice because it doesn't just judge based on rep

But there needs to be a way of alerting users on the site to flag the post

@GregP Why?

as the bot cannot delete it without review

I suck at farming rep, and I'd rather moderate then contrive ways to get rep to be able to do so properly

@undo as there are always those mistakes

@GregP Mistakes being?

false positives

the only way to uncover those is with human review

Its my biggest gripe with SE. Moving it back in means we're back to everything based on rep. All the flag queues, everything that SE does in terms of handing trust out, its all rep based. And then I'm hosed again. Maybe my personal feelings are irrelevant here, but thats what they are.

But this does not need to be confined to charcoal

I feel like I'm not communicating how I see this working: (1) Smokey detects a post. (2) We kick it into the spam queue, which is now shown to network moderators. (3) Record the result of that spam flag, use it as feedback.

It can be done by pinging active chat rooms (as it is now) and then if the post is deleted that will marry up with the notifications.

so there's no need to give active feedback to the bot, it can be done by posts deleted or not

@GregP this

@GregP but not this

@undo I don't think that is a good idea

@GregP Why?

I think it needs to be feed to all users (with some minimum rep constraint)

Why?

(trying to get to the bottom of this)

Because it will then all be put on mods

@GregP So?

We already handle a lot of stuff. This'd be a drop in the bucket, and these are the easiest drops to handle.

If you want just mods to handle all of this, why not make smokey mod only?

because as it stands the mods are not the one's deleting the spam, the community does

@Magisch Because we're not set up for that yet. We don't have a way to automatically funnel stuff into the queue.

So thats the plan? Make smokey moderator automated?

@undo that's not the impression I've been given of mods and sparing them too many unnecessary flags

@Magisch That's my plan

or at least my idea

That means there isn't a place in that project for me then.

I don't see why the community cannot handle spam and abusive flags, it would spare and share the load

fair enough, I suppose

I'm not going to force it on anyone

@Magisch nor most of us

But that's not the point

I think the more things that can be spread across the wider community (successfully) the better

Remember, a single moderator can nuke + destroy in seconds. That's far, far preferable to having 6 folks spend time on it.

Spam volume is... remarkably low, really.

well if you guys are going to be around supervising the spam 24/7

I don't see you guys deleting it on the spot and there's plenty of you around

You'll need to be live pinged and pay attention to it

I was actually pretty happy to have found a way outside of CVs where I could make a difference ... but fair enough, I suppose

go there immediately for every spam ping

Its more the small sites that have long lived spam

@GregP Spam? Yeah, we kinda do. I'm looking at SO's queue right now. There are two spam flags, from three and five minutes ago.

Well then the mods on all the sites will have to do that, which means giving power across sites that is not going to be well received by all sites

And those are comparatively old. Spam flags are super easy.

spam can sit on the sites for a long time

@GregP And that's the hardest part of this. It can be solved.

not so much SO, as it has a lot of users

@undo not really, you are trying to push the responsibility onto a very small bunch of users.

As it is there's a whole pool in socvr that jumps on spam in SO

and you're cutting that out and putting it on a handful of people

and there's time there's very few mods around

they're asleep

@undo So all we do here ... is just a band aid until you can find a way to let the real trusted users handle it instead?

Very few around in chat.

[ SmokeDetector | MS ] Blacklisted user: how to pass arguments in java main method any time when the application is running is this possible in any way how by Sameer Agarwal on stackoverflow.com

@undo Hmm, not sure that network mods would really want the burden of that. But perhaps in terms of a network mod spam flag equaling a mod spam flag or something like that.

@hichris123 That

And what you're doing is expecting the mods from the small sites to oversee the spam on SO also

a huge job

@Magisch I'd like to streamline this as much as possible. This is a necessary step, but the more steps we can eliminate the better.

@GregP Never said that.

@undo You want only the mods to handle it

So which mods handle what then? they stick only to their own sites?

This really is quite illogical undo, I have to say.

I was hoping we could ... be a part of handling this stuff more efficiently.

Remember, I'm not pushing this on anyone.

I'm not going to force it.

This was one of my main fears when they started talking about integration, that it will go back to all mod based and we'll be obsolete

You're a lone ranger on this one

Maybe.

Heck, I'm probably wrong.

I do think we can get rid of feedback.

I think it's a fair idea. Or integrating it with SpamRam in some capacity.

I do think we could get rid of 6-flagging

If you have a percentage of users who are active in moderating the site, let them

that's a separate issue and I don't think they will reduce that lightly (the min 6)

I'm disappointed. Not surprised, but disappointed nonetheless

making smokey mod only will mean the update speed of its patterns will be slower

There's no issue with flagging if enough people are alerted

@Magisch don't be, it will not take off, they won't cope.