Conversation

Notices

1. :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy: (kaniini@pleroma.site)'s status on Tuesday, 02-Jul-2019 03:26:45 EDT :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy:
   anyway, the solution for the womenarestupid.site spam is to look at the behavioural side of the accounts.
   1. :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy: (kaniini@pleroma.site)'s status on Tuesday, 02-Jul-2019 03:27:32 EDT :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy:

      in reply to
      generally speaking, there isn't enough useful data in the message content itself to make a determination. the *metadata* however, tells a different story.
      1. :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy: (kaniini@pleroma.site)'s status on Tuesday, 02-Jul-2019 03:31:36 EDT :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy:

         in reply to
         the mitigation I included in Pleroma 1.0 was too naive and was undermined by people using the sharing features (likes and repeats) to spread the spambot's initial post.
         1. :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy: (kaniini@pleroma.site)'s status on Tuesday, 02-Jul-2019 03:39:00 EDT :abunhdhappyhop: :abunhdhappy: :abunhdhop: :abunhd: :abunhdhappyhop: :abunhdhappy:

            in reply to
            we have options though.
            
            one is to extract the list of URLs and check them against a domain DNSBL like URIBL.
            
            another is to more carefully track each account's activities and use it to train the classifier. instead of allowing *any* previous accepted message, only allow messages that do not incentivize undermining as an initial message. this one is trickier though.