On Mon, Jan 01, 2007 at 09:13:06PM -0500, Kai Raab wrote: > > I've been running Dspam for a couple of weeks now. For users that > were previously receiving regular spam, I have found that within the > first day or two of training, the filter already catches over 95% of > spam w/ no false positives... except for one user. After 1 week's > worth of training, Dspam is only quarantining roughly 25% of his spam. > It even continues to deliver as innocent items tagged Resend, which > were marked earlier as spam. > > After a few days of poor results, I put his statisticalSedation value > down to 1, but saw no noticeable change. > > My inclination is to wipe his training history and start from scratch. > Does this make sense? And if I use dspam_clean for this, can someone > suggest the right syntax?
This is probably a good idea. It's possible that he deleted a spam or two instead of retraining and while I guess dspam will eventually get over it, I imagine having the spam tokens scored as innocent is going to confuse it a little. I don't think dspam_clean can be used to purge users completely. If you're using a database backend, probably the best bet is to delete anything from the dspam_token_data table which has this user's userid. The userid can be found in the dspam_virtual_uids table; or the actual system uid if you're using real accounts. You should also remove his data from dspam_stats as well, so the numbers make sense. Signature data can be kept or deleted as you wish; it will be deleted at some point by the periodic dspam_clean or similar maintenance we assume you're running.
