Conversation

Notices

1. Embed this notice
   jwz (jwz@mastodon.social)'s status on Wednesday, 20-Dec-2023 07:21:50 JST jwz

   "I got 3 whole spam messages today, so I'm going to defederate from a server that comprises over 1/5th of the entire network" is certainly a choice. That some admins are making today. Good luck with that.

   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Wednesday, 20-Dec-2023 07:21:30 JST Evan Prodromou

     in reply to

     • jaz :twt: :wales_flag:
     • Luis Villa
     • Simon Frankau
     • IFTAS

     @luis_in_brief @sgf so, @jaz and the folks at @iftas are doing great work on tools for moderation. For the very specific question of spam filtering, I do not know if they are pursuing it, but I very much believe that simple models like Naive Bayesian filtering, so successful for email, can be a big help.

   • Embed this notice
     Luis Villa (luis_in_brief@social.coop)'s status on Wednesday, 20-Dec-2023 07:21:35 JST Luis Villa

     in reply to

     • Evan Prodromou
     • Simon Frankau

     @sgf Yeah, I guess the question is “who is working on it and how can we fund that”. Might as well start now. @evan do you know?

   • Embed this notice
     Simon Frankau (sgf@mastodon.xyz)'s status on Wednesday, 20-Dec-2023 07:21:39 JST Simon Frankau

     in reply to

     • Luis Villa

     @luis_in_brief Yes.

     I think you can make the distinction that Threads might force the issue when otherwise the tooling could grow organically over more time, but given that Masto dev has historically left things until crisis point was reached (AIUI, IIRC, reading how e.g. instance blocks were developed), maybe forcing the issue is no bad thing.

   • Embed this notice
     Luis Villa (luis_in_brief@social.coop)'s status on Wednesday, 20-Dec-2023 07:21:43 JST Luis Villa

     in reply to

     • Simon Frankau

     @sgf @jwz Yeah, lots of the discussion about the hypothetical coming wave of Threads spam and harassment seems to ignore that any conceivable future for Mastodon must involve sophisticated T&S tooling and someone needs to write that, Threads or no Threads.

   • Embed this notice
     Simon Frankau (sgf@mastodon.xyz)'s status on Wednesday, 20-Dec-2023 07:21:45 JST Simon Frankau

     in reply to

     • Luis Villa

     @luis_in_brief @jwz Ironically, the incentives are in the right place for the wrong reason.

     Appropriate anti-spam tooling should be built into the Mastodon software, not something that large instance owners should have to roll-their-own.

     Fortunately, the owners of mastodon.social are also the Masto devs, so it all works out!

   • Embed this notice
     Luis Villa (luis_in_brief@social.coop)'s status on Wednesday, 20-Dec-2023 07:21:49 JST Luis Villa

     in reply to

     @jwz I’m not sure it’s a great idea (especially when “stay federated but block all DMs to non-followers” is apparently an implemented option?) but “if you can’t deal with your spam, your users will pay a price and complain to you” seems to help get incentives in the right place?

   • Embed this notice
     Evan Prodromou (evan@cosocial.ca)'s status on Wednesday, 20-Dec-2023 12:20:43 JST Evan Prodromou

     in reply to

     • Emelia 👸🏻
     • Luis Villa
     • Simon Frankau
     • Jaz (IFTAS)

     @thisismissem @jaz @luis_in_brief @sgf exactly, and also when other servers receive it

   • Embed this notice
     Emelia 👸🏻 (thisismissem@hachyderm.io)'s status on Wednesday, 20-Dec-2023 12:20:44 JST Emelia 👸🏻

     in reply to

     • Luis Villa
     • Evan Prodromou
     • Simon Frankau
     • Jaz (IFTAS)

     @jaz @evan @luis_in_brief @sgf for account creation, we do get the IP address and email, so yeah, we could maybe do spam scoring there. But yeah, overall spam needs to be handled at creation point (e.g., it's possible they'd be buying “clean" accounts and then using those to spam, so you want to catch at time of posting a status, as to block it)

   • Embed this notice
     Jaz (IFTAS) (jaz@mastodon.iftas.org)'s status on Wednesday, 20-Dec-2023 12:20:45 JST Jaz (IFTAS)

     in reply to

     • Emelia 👸🏻
     • Luis Villa
     • Evan Prodromou
     • Simon Frankau

     @thisismissem @evan @luis_in_brief @sgf I'm thinking we can be helpful mitigating spam account creation, I'm not sure how helpful a third-party can be with spam content classification in as timely a fashion as would be needed to combat spam messaging. Still, happy to discuss and support any and all activity that moves the needle.

   • Embed this notice
     Emelia 👸🏻 (thisismissem@hachyderm.io)'s status on Wednesday, 20-Dec-2023 12:20:46 JST Emelia 👸🏻

     in reply to

     • Luis Villa
     • Evan Prodromou
     • Simon Frankau
     • Jaz (IFTAS)

     @jaz @evan @luis_in_brief @sgf another thing I _really_ want in the admin UI's from reports is the ability to “submit" the report (sans comment) to an upstream server such that we can pool data on spam, harassment, and illegal content; this would allow multiple instances to participate together to train a shared spam filter model that can then be deployed within their infrastructure.

   • Embed this notice
     Emelia 👸🏻 (thisismissem@hachyderm.io)'s status on Wednesday, 20-Dec-2023 12:20:47 JST Emelia 👸🏻

     in reply to

     • Luis Villa
     • Evan Prodromou
     • Simon Frankau
     • Jaz (IFTAS)

     @jaz @evan @luis_in_brief @sgf The status.created/status.updated webhooks could in theory be used for spam detection & reporting too, but the post by that point would already be out in the fediverse, so you'd already be getting anger from people for "distributing spam".

     You really need spam detection built-into the software, much like Pixelfed has, which detects spam as it's created based on naive bayesian filtering that's been trained on known spam.

   • Embed this notice
     Emelia 👸🏻 (thisismissem@hachyderm.io)'s status on Wednesday, 20-Dec-2023 12:20:49 JST Emelia 👸🏻

     in reply to

     • Luis Villa
     • Evan Prodromou
     • Simon Frankau
     • Jaz (IFTAS)

     @jaz @evan @luis_in_brief @sgf the infrastructure we're standing up for CSAM scanning could in theory be used for spam detection, but tbh, that's probably too late — ideally even with CSAM it'd be before the post is actually created & federated, but we don't have that hook so it has to be afterwards & the scanning process takes some amount of time.

   • Embed this notice
     Jaz (IFTAS) (jaz@mastodon.iftas.org)'s status on Wednesday, 20-Dec-2023 12:20:50 JST Jaz (IFTAS)

     in reply to

     • Luis Villa
     • Evan Prodromou
     • Simon Frankau

     @evan @luis_in_brief @sgf Spam is on the list thanks to the needs assessment, but it's not on the front burner. Here's our current activities (not great on mobile, sorry):

     https://cryptpad.fr/kanban/#/2/kanban/view/E4DOgQKxzlKDmVescviFZRXS8L0MM-JNGxIoo20sk2M/

     Happy to talk with folks if we can support any activity in this space right now, otherwise we'll be looking early '24 to see if there's APIs or services we can help connect or acquire to help remediate the inevitable spam.