More on spambayes...

[alexpgp]

My subjective feeling regarding the spambayes spam-filtering software is that it is working, and working better as time goes on. Interestingly enough, a lot of the Nigerian 419 scam letters keep getting through, but that's probably the result of the letters looking so much like legitimate e-mail.

In reviewing incoming spam caught by spambayes, here's a breakdown of how many messages were caught by the spambayes, out of every 10 consecutive spam messages, for the first 300 messages in the file (arranged chronologically):

Messages	# caught	% of spam caught, total	% of spam caught, last 50 msgs
000-010	3	30	-
011-020	6	45	-
021-030	7	53	-
031-040	5	53	-
041-050	2	46	44
051-060	6	48	52
061-070	8	53	56
071-080	5	53	52
081-090	3	50	48
091-100	6	51	56
101-110	5	50	54
111-120	7	52	52
121-130	7	53	56
131-140	4	52	58
141-150	9	55	64
151-160	10	58	74
161-170	6	58	72
171-180	4	57	66
181-190	5	56	68
191-200	8	58	66
201-210	7	58	60
211-220	9	60	66
221-230	6	60	70
231-240	6	60	72
241-250	4	59	64
251-260	7	59	64
261-270	9	60	64
271-280	8	61	68
281-290	9	62	74
291-300	7	62	80

The "dips" correspond to when I check my mail and "train" spambayes by passing a message through its analysis engine, after which it is sent to the spam file.

I get the impression that there is an inexorable creep upward in the trend of spam messages caught by the software. Indeed, I'm noticing that I'm seeing (though I may not actually be receiving) a lot less spam than I used to, and that my main bugaboo right now is messages that aren't spam, but aren't terribly welcome, either. (Just call me Mr. Fussbudget.)

Cheers...

Comments

[bandicoot.livejournal.com]

As to the last, the perfect solution already exists, but unfortunately the phone companies frown on sending 20kv through their lines ;)