From 61836a7df0929ad78c06dbd1e39d8455dea01d0f Mon Sep 17 00:00:00 2001
From: Gustav Eek <gustav@fripost.org>
Date: Sun, 18 Sep 2016 16:28:18 +0200
Subject: Questions on spam logged in ideas and faq

---
 faq.mdwn | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

(limited to 'faq.mdwn')

diff --git a/faq.mdwn b/faq.mdwn
index 7ae4748..a10bfd6 100644
--- a/faq.mdwn
+++ b/faq.mdwn
@@ -233,6 +233,27 @@ du skriver brev.
 Se även [Sender identities](https://docs.roundcube.net/doc/help/1.1/en_US/settings/identities.html)
 på [Roundcubes hjälpsidor](https://docs.roundcube.net/doc/help/1.1/en_US/).
 
+### Hur slipper jag skräppost? 
+
+> Spam-filter, finns det något jag kan hjälpa till för att slippa
+> några av de spam som når min inbox?
+
+Administratörens svar på engelska följer:
+
+> A general issue with Bayesian spam filters is the privacy implication:
+> the filter has to read each message, digest it, and store a token in a
+> database.  For this reason spam filtering at Fripost will always be
+> opt-in.  Another issue is the training: for obvious privacy reasons we
+> want per-user training, not global training, so each user would have to
+> train their filter themselves.  In particular, bootstrapping a Bayesian
+> filter requires feeding it with a generous sample of spam and ham
+> messages (200 each for spamassassin, IIRC).
+> 
+> While the technology is already in place, there is currently no way for
+> users to opt-in, tune their preferences, or submit sample messages :-/
+
+Se mer om detta i [[hjalpa_till]].
+
 ### Misstänkta brev -- Skickar Fripost ut skräppost? ###
 
 > Jag har fått ett mycket misstänkt brev från någon med ett
-- 
cgit v1.2.3