Add settings to auto-learn ham or spam at certian thresholds

[jwgreene@…]

Add a setting that does the following:

If Quarantined message is at a certian score or higher than auto-learn it as spam so the user does not have to click spam. If a message in Ham Cache is a certian score or higher auto-learn it as ham so the user does not have to click ham.

While this could cause problems we are finding that many users want a system where they do not have to do anything and want the system to do it for them.

[dmorton]

Unfortunately, it is possible for a system to drift, and this could cause greater problems if reporting is turned on.

My original installation at my ISP is showing some weird effects - some spam is coming through as -4.9 score.

[rjl]

More seriously, there are reporting issues to deal with. Generally, spam reports are only supposed to be done for items that have been confirmed by a human being, so "auto-confirming" items that score above a certain threshold is potentially dangerous.

More specifically, Razor's usage terms expressly forbid the reporting of items without human confirmation: "Only spam detected by humans should be reported to Razor. Automata, however good it is at detecting spam, isn't human, and should not be allowed to make a report or revoke decision. The only exception to this rule is troll addresses that have been seeded for the purpose of gathering spam." [Razor FAQ]

On the other hand, it should be possible to auto-*delete* items that score higher than some threshold, which is certainly safe, but has the downside of not being able to use those items for learning or reporting purposes. With certain flags set, perhaps, learning could be divorced from reporting in this case, so that such items could still be used to train the Bayes database, even if they're not reported.

In your scenario, Jim, where some of your users don't want to manage their own quarantines and caches, it might simply be best to remove their Maia accounts altogether and let their mail end up in your domain-default or system-default pool, for an administrator to manage for them.

[anonymous]

Once the administrative views are complete, admins should be able to manage spam for all accounts if needed. Having all the queues in one place will be great!

[maia@…]

I am finding increasing presure from my users that they don't want to have to confirm mail all the time. They just want it to work and if they are missing mail that they can check the quarantine for it. They also want stats for thier own domains. This rules out using a sitewide quarantine. I think having auto-confirm is a must now and maybee we could have any mail AUTO CONFIRMED set to NOT REPORT to third parties. I have been hacking up a cron script that does jst what I want. If I get it running I will let people know.

Regards lunix_aus ( Mick Pollard )

[dmorton]

Mike: It is possible to turn on auto learning in SpamAssassin, so that all mail over or under certain scores get learned automaticlly. It still stores the messages in Maia as usual, so they can still be corrected if there was a mistake.

Also, in the stats bracnh (to be 1.1) there is support for stats collected per domain. In addition, it has support for a "KILL" level where messages would be simply dropped out of Maia for going over a specified (high) score.

The only reason for keeping this ticket around is if we wanted to make these settings available in a web interface rather than in the SA config file.

[rjl]

This is the same issue being addressed in #437.