Spam prevention - algorithm block more posts than before

Hey ​@revote 

Not really, the spam markings we are getting seems to be aligned with what we have in our settings.

Just a thought (you probably already checked this) - you could revisit the keyword blocker and see if any words added there is causing this change? 

We check this every time and no, our members discuss nicely (almost) all the time.

As said, this is just a hunch. It would be nice to hear more experiences.

Hmm interesting - following the thread for other’s experiences.

Is there anyone who see normal and regular posts be marked as a spam, more often than before?

Is it possible to teach algorithm, if/when we mark posts not_spam?

This is difficult because we dont know why post was marked as a spam and we dont know to do differently in the future.

If this continues I have to think about to disable pre moderation.

We’ve seen normal posts being marked as spam, but all of our examples have urls in the content.

We have links also but this isn´t the situation with all of the cases 🤔

--

Have you think about, that every time platform hides the post, it is DSA case and we have to take needed actions?

So, the more the platform hides posts, the more it means work for us. For vain.

We’ve seen this as well, but figured out it was somewhat self-imposed. Earlier this year we started seeing more posts that looked normal (usually AI generated) but contained parasitic SEO or malicious links. Logically we flagged them as spam, but it turned out the spam filter wasn’t picking up the subtle nuances and started seeing normal posts as potentially spam.

We haven’t been able to undo it entirely, but my entire team knows now to only flag super obvious spam as such (and ban the author) and for more normal looking spam posts to ban author and trash without flagging as spam.

We’ve also just started a pilot to plug in AI detection into our moderation workflow, so that should help us identify more spam without tripping the spam flag right away.

Oh, I would love to hear more about this?

​@revote I’ll share a more comprehensive post once we’ve been able to pilot for a few weeks, but the short version is that we’re using Zapier (automation), Winston AI (linguistic modeling), and Slack (alerts) to automate a review of all new topics and replies for URLs and AI. This also gives us a solid foundation to add other spam detection such as keywords and maybe even sentiment.

​@revote so do you get a lot of content that is marked as spam and stays hidden?

I can´t say a lot, but the hunch is that more often than before 😊

This topic came to my mind, when we saw today two posts going to spam with no reason. No urls, no swearwords, no nothing. I think there’s something to this 🤔

Hi Suvi and everyone, this can happen - an automated system like this will get it wrong sometimes, and there’s no easy way to determine exactly what’s going on under the hood.  The best action to take is to simply mark as ‘not spam’ as the algorithm will learn from this, and will hopefully adjust to not make the same error again.  If it seems like it’s happening more than normal and is causing concern I do recommend creating a ticket with our Support team as they can dig a bit more deeply and can relay feedback to Akismet (our 3rd party spam fighting algorithm provider) if needed.  

Thanks ​@Kenneth R.

I am sorry but I and we have feeling that it doesn´t learn anything 😀

I think I have created a ticket (Maybe once or twice? If I remember correctly, hopefully!), and I have got response like “there is nothing we can do”.

And this is why this is difficult case - active or superusers are annoyed and we can´t do anything for it. I can only disable pre moderation.

​@revote It certainly is hard to know for sure if or when it is learning.  

There’s a practice that I often recommend, and you’ve probably done or thought of doing it yourself already, but if you have active members who are frustrated by false positives, you can create a custom role that exempts members from the spam checker and auto-assign that role relatively early in the rank structure.  That way, anyone who has at least reached a certain minimal threshold where you’re fairly confident you would have spotted them as a spammer by then, will always be exempt.

Thanks, I didn´t know about this. I have done it now to our community. Although this isnt the best solution, I think this was ok workaround.

Btw, today one post from our moderator was marked as a spam. User with Moderator primary role.

Hey ​@revote - could you PM me the link to the topic where your Moderator got a post flagged as spam?

I can, but I asked this from Support and they said that Moderators & Community Managers & Admins -roles are not bypassed from spam filter. They didnt know why this is so.

They should be bypassed, so I will raise an idea about this.

Hey ​@revote - yes I heard that from them too, and I’ve pinged our Product team as I also find it really strange if that’s the case.

Hi ​@revote - our Product team just confirmed to me that Admins, CMs and Mods should indeed be exempt from the spam checker.  What you encountered is a bug, and we’re fixing it.

Latest case:

User posted to our community. Long post. After publishing, user noticed that there were typo, so user edited and fixed it. Then platform marked post as a spam.

Then user posted two more posts, to other topics. They were marked as a spam also.

--

All of those 3 posts are ok, we dont see any reasons why they were marked as a spam.

We have spoken with our moderators and there are other cases during latest months, when editing the post triggers spam filter and post will be hided. Is this planned behaviour?

Let´s assume, that editing the post, triggers spams filter and post will be marked as a spam. Okay. Why the next posts will be marked also as a spam?

I am sorry but this spam filter doesnt work, this is annoying.

LOL ​@revote - We have similar with both spam and pending folders. At least a few times a day something gets caught and often as you say it happens to the same person on a few of their posts for no apparent reason. I’ve kinda given up trying to figure most of them out since when I have tried, rarely can I see why. Unfortunately it happens to our power uses the most and they just “roll their eyes” now and wait for approval.

This occurred again, with other user.

If first post is marked as a spam, next ones will be also marked as a spam.

Is spam prevention still on in your community? Or have you created spam bypass role for certain users?

Hey ​@revote - I’ve not heard of the editing of posts being a factor, but let me check with the teams about that. 

And, indeed, ​@Chris Hackett - as Revote says it’s a great idea to exempt your most active users from the spam checker.  You can do that easily with a custom role, and could even assign that role automatically as part of your rank structure so that everyone apart from new-ish members is exempted.