Quantcast
Channel: Can a machine be taught to flag comments automatically? - Meta Stack Overflow
Viewing all articles
Browse latest Browse all 5

Answer by Tim Post for Can a machine be taught to flag comments automatically?

$
0
0

This post alone is an amazing amount of work, the amount of work it actually represents is admirable. It's not an odd itch to want to scratch, we've been putting a lot of thought and time into comments lately. While we did find a few nice things to say about how comments on our sites are actually used, the majority of the time we've spent has gone into talking about what we don't like, or basically the majority of our comment system and accompanying moderation tools.

Go back for a second on why we decided to implement them, which was a growing problem with litterbugs. As people on the Internet are so accustomed to being able to comment on practically everything, folks were using answers to write comments despite glaring UI fixtures asking them not to do so. Subsequently, Jeff cried caved, and we had comments, followed by a system that was designed to help elevate anything useful contained in them above the "+1 lol jQuery ftw" baseline.

10k users could also delete any comment in the system. We did our absolute best to make sure people didn't get too attached to comments because we correctly anticipated that which was inevitable, people were going to make a lot of noise using them. Mods were informed that mercy is completely optional when faced with cleaning up a noisy thread of comments, and users were warned that comments were at best ephemeral.

The noise comments generated soon surpassed the noise that we hoped they would keep out. We then put a rep requirement on them so that .. oh wait, yes, that's right, we ended up with:

  • People were using answers to write comments, so we should give them a place to write comments instead
  • We then restricted comments so that only people that know how our system works to some extent can use them
  • Many people that tend to write noisy comments come from searches, have no or little rep, so they tend to write answers inst...

Headdesk.

If you're ever a mayor of a city and find that you have a litter problem, we strongly recommend welding most of your trash cans shut; it's worked so well for us.

In the city analogy, you've found litter right next to a trash can, yet you have to call the sanitation department to send someone to come throw it in. That's not good for you, that's not good for them, in fact if it were any more inefficient we could make a board game out of it with the end goal being to trap mice. The work you've done and the effort is fantastic, but I think we're going about it the wrong way, and I think we've got to bite the bullet.

The chief problem with comments is that comment moderation tools stink.

That's right, they stink. Too many clicks, too many people involved, too much thinking and work over things that were never supposed to matter much to begin with. We've talked about algorithms to hide comments, different kinds of flagging tools, all kinds of stuff - here's what I think we need, and what I'm chewing on at the moment:

Just-in-time help

Because you know, please don't comment to show appreciation, click on the up-vote or (anon) helpful button instead. Comments that don't directly address points in the post just create more noise to read and are routinely removed. We have this ability now, sort of, and can probably expand it. Basically, the same matches you outlined.

Badge-based trust

Enhanced privileges for gold tag badge holders has worked out extremely well. In fact, I'm planning to expand that idea next year. Why can't we let folks with flag-based badges just delete noise when they see it?

I think we've complicated what was initially a very simple system enough, and need to get back to basics. No, I don't want any 10k user to be able to whack things from orbit on a whim, but I tend to trust those that our moderators trust.

There are issues and caveats, the chief one being that people tend to be very attached to everything they type on the Internet. But when you warn them based on what they're typing .. that's teaching.

I've got to chew on this a bit more, but this was a wake-up call. We've got to address the issue sanely, or better mousetraps start looking dangerously attractive as time goes on.


Viewing all articles
Browse latest Browse all 5

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>