It is tempting to avoid filtering outbound mail (with SA or other). I am
assuming that outbound mail is legitimate (users are honest, and logs
can be used to look for abnormal behaviour and punish the guilty).
Now my question. Wouldn't that weaken Bayes filtering? I see two views:
- no: after all, The Bayes engine needs to learn inbound mail since
that's what it will be filtering.
- yes: if it checks outbound mail, the Bayes engine will learn
words/tokens that are legitimate, and will thus be less FP-prone.
In the latter view, one can still feed outbound mail to SA for learning
only. However, would there be any benefit in this compared to just
filtering the mail?
Similarly, what would be the effect on AWL?