Dnia 7.08.2021 o godz. 10:18:07 Philip Paeps via mailop pisze: > >up to the point where it is impossible to manage their spam > >filtering correctly > > Unfortunately, "correctly" is always going to be subjective. Some > users subscribe to newsletters, {foo}-of-the-day, discount codes and > all manner of other stuff. Other users get deeply upset if x hours > after buying something and getting a receipt, they get a message > suggesting they might want to buy something else.
Therefore, apart of obvious cases to block (widespread distribution of the same, unsolicited email to random users, obvious scam/phish or messages containing known malware) the spam filtering needs to be adjusted per user (Bayesian filtering is pretty good at this). That's exactly what I mean by "correctly". But this doesn't scale well, both in terms of computing power needed for content filtering and human work needed to manually adjust filters whenever needed. Small providers, who have a few thousand users, can do this, but the big ones, who have millions of users, can't. And the first goal should be always to minimize the amount of false positives, ie. lose as little legitimate email as possible, while filtering as many spams as possible is a second goal. In case of doubt, first goal should always win, ie. the message should be delivered to recipient's mailbox. If the recipient flags it as spam afterwards, then further messages similar to this one can be filtered (but for this recipient only, not for other recipients, with exception of the case when many recipients - say 100 - mark the message as spam), but there should NEVER be filtering based on pre-assumptions, like "this IP is in a bad neighbourhood, so although it never actually sent spam, don't trust it by default and mark messages from it as spam". In short, the spam filtering policy should be "default allow unless confirmed as spam" - and what exactly "confirmed as spam" means must be continually adjusted over time, in large part by human intervention of the admin; you cannot fully "outsource" this task to any automated solution, even Bayesian filters, because - as it was already said multiple times on this list - the users may not understand correctly what flagging a message as spam means. (For example, if some recipient marks purely transactional email as spam, understanding spam as "I don't need it", the admin should anyway whitelist these messages so they still get delivered). I'd say, it's a kind of an art, and it requires staff dedicated for this task. Again, this doesn't scale well, and that's the reason why big providers can't do this properly. Too bad that people who sign up for a mail account at a big provider are unaware of this. They are also unaware that there may be false positives at all, and therefore they *should* look regularly into their "spam" folder to check if there aren't any legitimate messages there. Nobody (especially the provider) tells them this clearly; they think (are made to think?) exactly the opposite - that some mysterious AI filters the spam for them better than they could themselves and if something gets into "spam" folder there's absolutely no need to look at it, because it's certainly something they don't want or need. I'm not sure if this is a problem that can be solved by legal intervention, but I'm sure that besides legal intervention there are no other means that could solve it. In other words, if legal intervention doesn't help, nothing helps. -- Regards, Jaroslaw Rafa r...@rafa.eu.org -- "In a million years, when kids go to school, they're gonna know: once there was a Hushpuppy, and she lived with her daddy in the Bathtub." _______________________________________________ mailop mailing list mailop@mailop.org https://list.mailop.org/listinfo/mailop