search

Charcoal-SE / SmokeDetector

Headless chatbot that detects spam and posts links to it in chatrooms for quick deletion.
https://metasmoke.erwaysoftware.com
Apache License 2.0
474 stars 182 forks source link

Manually reported question: Show which rules would have hit #513

Closed tripleee closed 6 years ago

tripleee commented 7 years ago

When a post has been manually reported, it's hard manual work to figure out if it was reported because Smoke Detector missed it (which is often a problem which we want to spend time on solving) or just because the reporter was quicker than Smokey (which of course isn't a problem at all). Could the analysis be run on these questions after they've been reported, and included in the Metasmoke record, so that it's easier to see when we have missed something?

tripleee commented 7 years ago

Really a feature request for Metasmoke actually, I guess. Is there a way to move it?

Undo1 commented 7 years ago

I'd say it should stay here, actually - SmokeDetector has all the parsing code; this is something that would be implemented here, not in metasmoke.

ByteCommander commented 6 years ago

I'd like to give this issue a little bump...

superplane39 commented 6 years ago

What would happen with the weight of the !!/report command with this? Could this theoretically result in manually reported post being autoflagged?

angussidney commented 6 years ago

Yes. However, if the post hits our checks it should have been reported in the first place, which means that it would have been autoflagged anyway.

The only issue is that the 'Manually Reported <Q/A>' reason itself could take a 1-reason post (~90 weight) to 2 reasons, bringing it within low autoflagging territory (~170ish). We can always decide later whether we want the extra human intervention to count as extra weight.

iBug commented 6 years ago

https://chat.stackexchange.com/transcript/message/42954284#42954284

Maybe implement a !!/scan <URL> command to force a scan on a given post, and nothing should happen if the post is all right.

Edit: Check #1634 I made an attempt do implement one, someone test if it works and see if there's room for improvement.

iBug commented 6 years ago

1634 is a good solution for this and it does all the job I wanted: Scan the post and report it, and if the post is all right then do nothing (it does reply Post does not look like spam as a clearer indication). It's rolled out and (hopefulley) good to use now.

Edit 2: Finally implemented this FR in #1711 . Should be very good now.