[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Ifile-discuss] Effect of widely differing volumes on ifile classificati
From: |
Jack Bertram |
Subject: |
[Ifile-discuss] Effect of widely differing volumes on ifile classification |
Date: |
Thu, 20 Mar 2003 10:10:41 +0000 |
User-agent: |
Mutt/1.4i |
I use ifile to classify all my emails, including the mailing lists I
subscribe to, and I tend to get a classification success rate of around
95-98%.
Recently, the rate, which had been consistent for some time, began to
plunge to about 50% and stayed there, until I deleted .idata and rebuilt
it from scratch, and it's now classifying better than before. (Data
attached at bottom for completeness)
I'm wondering whether ifile's database got "overloaded" by having large
numbers of different volumes sent into different categories. However,
if that were the case, then I'd have expected it to start classifying
emails into the mailing list categories rather than into other
categories. This didn't happen - it actually started to misclassify the
mailing lists which receive all the volume.
Any ideas what could cause this "chaotic" effect?
jack
Date Total Incorrect Percentage
---- ------- --------- ----------
Thu Feb 20 160 8 95%
Fri Feb 21 189 7 96.2%
Sat Feb 22 164 2 98.7%
Sun Feb 23 45 3 93.3%
Mon Feb 24 85 1 98.8%
Tue Feb 25 136 2 98.5%
Wed Feb 26 171 7 95.9%
Thu Feb 27 167 23 86.2%
Fri Feb 28 164 44 73.1%
Sat Mar 1 166 58 65%
Sun Mar 2 131 29 77.8%
Mon Mar 3 129 17 86.8%
Tue Mar 4 195 43 77.9%
Wed Mar 5 279 62 77.7%
Thu Mar 6 180 50 72.2%
Fri Mar 7 257 9 96.4%
Sat Mar 8 157 28 82.1%
Sun Mar 9 106 0 100% - didn't read email; no
reclassification done
Mon Mar 10 93 48 48.3%
Tue Mar 11 172 76 55.8%
Wed Mar 12 160 51 68.1%
Thu Mar 13 131 4 96.9% - regenerated .idata
Fri Mar 14 176 0 100%
Sat Mar 15 141 7 95%
Sun Mar 16 86 2 97.6%
Mon Mar 17 124 1 99.1%
Tue Mar 18 204 4 98%
Wed Mar 19 230 3 98.6%
Thu Mar 20 204 4 98%
- [Ifile-discuss] Effect of widely differing volumes on ifile classification,
Jack Bertram <=