forum - spam detection/processing needs improvement
- /e/ version: N/A
- Device model(s): N/A
Summary
The rules implemented in the forum to automatically detect spam and the associated automated actions need improvement.
The problem
See for instance these posts in the formum. Two people reporting similar false positives by the spam detection making all of their posts with links in them hidden causing frustration on the side of the posters and additional work for the forum admins/moderators.
Well, in my case, it seems that the problems began when I was attempting to post a reply containing a link to codeberg-dot-org. During editing (or was it when I hit the "answer" button, I don't remember) I was informed by a pop-up that links to codeberg-dot-org are forbidden by forum rules (for whatever reason... maybe a misconfiguration; at least I see other posts to that site). So I converted the link to just an URL text. Apparently the spam detection mechanism was not happy about that and decided that I was spamming the forum with forbidden links.
Within minutes all of my previous posts with links in them got hidden. I got several private messages that these posts were hidden as "the community feels it as an advertisement". Apparently the spam detection did not evaluate these other posts before deciding (in the name of the community) that they are advertisements.
Funny side note: one of the nonsense I was requested by the "discobot" to post during the initial tutorial in order to earn some more posting permission even got removed completely...
Furthermore, apparently I lost all my hard-earned badges I got during the tedious "discobot" tasks. For instance, I only just recently re-earned my "first link" badge...
I should add that I completely support the officially documented forum rules. However, I have yet to see a post of mine which was violating them.
What is the current behavior?
Spam detection sometimes reports false positives and causes collateral damage.
What is the expected correct behavior?
False positives cannot reliably be prevented, I am aware of that -- unless you make the detection more tolerant which will most certainly lead to significantly more false negatives.
But knowing that there might be false positives, please limit automated actions to just the post which allegedly contains spam.
And in case the initial false positive was caused by a misconfiguration of the detection rules (you really do not want to block codeberg-dot-org links, do you?): please find a way to test these rules first before arming them.
Technical informations
See the above mentioned complaints in the forum. In the follow-ups, there are a few links to posts which got automatically hidden.