[Spambayes] Early Experiences with SpamBayes

Leue, William M (Research) leue at crd.ge.com
Mon Jun 30 13:51:21 EDT 2003

Our large company uses Windows 2000 and Outlook 2000. After discovering
SpamBayes in an article in InfoWorld, I and 6 or so colleagues installed
SpamBayes. We have been testing it for about 4 weeks. Our experience is so
far mainly very good, but there have been a few glitches.

Installation (using the binary installer) was quick, easy, and problem-free.
Ditto setup and training. In the first couple of weeks, everyone testing it
had uniformly excellent results, even when the training set was rather
small. Incremental training using 'Delete as Spam' and 'Recover from Spam'
also seemed to work very well.

However, in the past 2 weeks, we have begun to see problems. The first of
these problems is I think the same as previously reported in the list,
"SpamBayes Disappears." As mentioned by Mark, it is not too hard to
re-enable SpamBayes when this happens. As a clue, the problem seems to
happen mainly on laptops, and it may be related to Outlook having problems
finding a mail server. This seems like mainly an Outlook problem to me.

The second problem that many people, including myself, have experienced is
that SpamBayes suddenly stops filtering, even though the control buttons are
still visible. There are two symptoms: first, filtering becomes ineffective
without warning, and spam suddenly starts coming through again.  Second, the
'Spam score' column that most of us added to our mail listing stops
appearing. These two effects are clearly related.

In a private email communication, I described this second problem to Mark,
and he said that the only recourse is to rebuild the database from scratch.
All of us who were having this problem did so, and rebuilding the database
did seem to fix the problem. Since it seems likely that the problem will
happen again, we are now all keeping some Spam and Ham folders around
permanently to rebuild the database should that be needed.

This second problem seems like a SpamBayes problem to me, although I don't
have any deep insight into it. It would be nice to have a diagnostic tool
that would allow us to view aspects of the database in a human-readable
manner, so that we could see if the database is hopelessly corrupted, or
whether the problem is somewhere else.

If this second problem could be isolated and fixed, it would be wonderful. I
know a lot more people in our shop would love to use it, but I hesitate to
recommend it to a wider audience in its current state. Has anyone else seen
this problem? Any ideas on how to fix it?

-Bill Leue

More information about the Spambayes mailing list