I’m going to propose a new moderation setting in mastodon.

Aleksandra Lesya

@jerry If only they could allow the Admin to have the same "filter" by word (like user have) but added server wide to cleanup some bad # like "# r*pe" and it's enforced for everyone.

@girlintech yeah, pleroma and Akkoma instances have MRF which allows for that. I think the mastodon team is philosophically opposed to that though

Emelia 👸🏻

@jerry have you seen my filters proposal? https://fires.fedimod.org/concepts/filters.html

It has a filter defined that's ignore-non-public but we could also have ignore-activities or something, that would achieve this (reject-activities would send a Reject activity back)

Renaud Chaput

@jerry @girlintech its not philosophical. This can very easily lead to legitimate posts being silently dropped (think different meanings in different languages for a word, partial word matching or not, blocking « a » by mistake…), without any user knowing about them.
We have not found a good way to handle this, and then decide on what to do if we present such a list. Should user be able to opt out the server filter? Individual posts? What about reply chains? And so on

Emelia 👸🏻

@renchap @jerry @girlintech one option might be to have a quarantine section for incoming activities, where they're held if they match a pattern and require manual review.

Aleksandra Lesya

@thisismissem @renchap @jerry or they could finally release a way to set a filter over one language to avoid the wrong language to be filtered by a bad filter.

feld

@thisismissem @jerry nooo please don't send anything back. This is a bad idea

Emelia 👸🏻

@feld @jerry there are cases where you want the server to know “hey, we rejected your message" but there's also cases where you’d want to silently drop the activities.

Emelia 👸🏻

@girlintech @renchap @jerry not all posts are going to be correctly categorized by language. Also, if you're receiving spam posts and start dropping “spam-keyword" for "english" the spammers pretty easily adapt by sending spam in "german" even though it's the same keyword.

For dealing with spam, we need naive-bayes based models, which can more accurately classify spam than simple keyword matches

Renaud Chaput

@thisismissem @jerry @girlintech Manual review by who? The admin/mods? This would be a huge burden, and what about "private" exchanges? Users? Do you really want to have them go through a list of messages that an admin decided were potentially very problematic? Reactions we got from the filtered notifications feature lead me to think that this will not fly well. Also should every user review the posts for their own timelines, meaning that a post might require a review by hundreds?

Emelia 👸🏻

@renchap @jerry @girlintech yeah, I'm not saying quarantine is the best solution, but it is a partial solution. But really we need proper naive-bayes based spam classifiers instead of simple keyword filters that drop stuff.

Aleksandra Lesya

@renchap @thisismissem @jerry in my case i live on my tough we need to copy the feature of bluesky.

List made by community, so user can decide to set them or not.

Unlike bluesky you could make them open tough

Renaud Chaput

@thisismissem @jerry @girlintech I have a feeling (and hope) that the first non-discovery FASP we work on will be this.

jenboonim 👻

@renchap @jerry @girlintech seconding this. There are hashtags I check regularly due to them being associated with bad content. However if I had blocked them entirely I would've been worse off because I would've missed IFTAS posts about how to protect against that content and users discussing which servers I need to block

Word filters are a poor substitute for moderation, particularly while Mastodon doesn't yet support a "filtered" state for posts where they're visible to mods but not users

Aleksandra Lesya

@renchap @jerry one thing could also warn user on a server when no mod have connected recently so they could move over a moderated server (you have added a way to block new inscription but warn old user can be good too).

Emelia 👸🏻

@renchap @jerry @girlintech I might also be able to do this by leveraging @MarcT0K's classifier: https://github.com/MarcT0K/Fediverse-Spam-Filtering/

feld

@thisismissem @jerry

> there are cases where you want the server to know “hey, we rejected your message"

That's what we thought about email too and it formed the basis for mail bomb / backscatter attacks. This is just the Fediverse making the same mistakes all over again...

I guarantee this will be abused and cause servers to have a massive backlog of Rejects in their queues. Especially the attacker can ensure the Reject can never be delivered successfully.

IMO every Reject activity should be from deliberate human interaction so it can't be weaponized.

Communicating state with Accept/Reject for Follow Requests is the only option we've got. But trying to further establish state across the Fediverse for other activities is not something I'd recommend.

We need to stop trying to make the Fediverse act like a centralized platform because it cannot and will not work that way.

edit: also the more we do things like this the harder it will be to self-host on commodity hardware as it continues to raise the hardware requirements to process the garbage that will constantly be trying to crush your little fedi server. We already suffer under Mastodon Deletes, please don't send Rejects.

@girlintech @renchap that is a good idea. I think the problem the mastodon team has is that they have far more good ideas than they have capacity to deliver, so we need to help support them

Emelia 👸🏻

@girlintech @renchap @jerry I'm also working on labelling as a moderation feature via the Social Web Community Group ActivityPub Trust and Safety Taskforce, which could enable things like this.

Bluesky Moderation Services are in a very similar implementation space to FASPs.

Renaud Chaput

@jerry @girlintech yes, thats a good idea, and should not be hard to implement. Need to think about what to display (ie a notification? A web banner? An email? A not-yet-designed-server-initiated-annoucement?)