管理员

Rank: 9 Rank: 9 Rank: 9

1^#

字体大小: tT

发表于 2009-5-8 10:04 | 只看该作者

[讨论] 收集DSPAM的使用效果

hi，all
调查的初衷是想知道使用 dspam 的用户群大概有多少，以及使用 dspam 至少半个月后的效果；
同时也方便大家一起交流，对社区以后是否改进dspam 的一个重要参考，请过路的朋友踊跃跟贴，谢谢大家。

请不要在这个贴子里发与dspam使用效果无关的信息，谢谢：

2010年1月12日22：00，100用户左右，使用Spam-lock+Dspam+Clamav,导入Extmail官方的训练库；
2010年1月13日15：30，识别率见下表：

Metric	Calculated as
Overall accuracy (since last reset)90.233%	(SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset)76.623%	(Spam catch rate only)
Spam ratio (of total processed)44.238%	Total SPAM messages (both caught & missed) / Total number of messages

	SPAM messages	Good messages
Since last reset	18 missed	3 missed
	59 caught	135 delivered
	76.623% caught	2.174% missed
Total processed by filter	481 missed	16 missed
Total processed by filter	2974 caught	4339 delivered
From corpus	493 fed	10 fed

2010年1月15日16：00 识别率

Metric	Calculated as
Overall accuracy (since last reset)81.81%	(SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset)74.04%	(Spam catch rate only)
Spam ratio (of total processed)50.01%	Total SPAM messages (both caught & missed) / Total number of messages

	SPAM messages	Good messages
Since last reset	413 missed	13 missed
	1178 caught	738 delivered
	74.041% caught	1.731% missed
Total processed by filter	877 missed	26 missed
Total processed by filter	4093 caught	4942 delivered
From corpus	1080 fed	13 fed

2010年1月17日统计一下24小时的投递状态“458 SPAM, 96 Good, 5 Spam Misses, 1 False Positives ”识别基本到了99%

2010年1月31日20：00，识别率
Metric Calculated as
Overall accuracy
(since last reset)99.02% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification
(since last reset)99.73% (Spam catch rate only)
Spam ratio
(of total processed)50.906% Total SPAM messages (both caught & missed) / Total number of messages
　
　 SPAM messages Good messages
Since last reset 1 missed 10 missed
370 caught 721 delivered
99.730% caught 1.368% missed
Total processed by filter 954 missed 241 missed
7953 caught 8349 delivered
From corpus 1222 fed 121 fed

收藏分享评分

ExtMail邮件开发网
liushaobo@extmail.org
探索高性能的Anti-Spam组合

回复引用

订阅 TOP

jgh2008

中级会员

Rank: 3 Rank: 3

2^#

发表于 2009-5-8 10:24 | 只看该作者

难得沙发：

以下是4－5天时间的效果统计：
                                 SPAM messages    Good messages
Since last reset                      31 missed             2 missed
                                             2110 caught          4215 delivered
                                             98.552% caught    0.047% missed

回复引用

TOP

808kkk1984

新手上路

Rank: 1

3^#

发表于 2009-5-8 10:43 | 只看该作者

Since last reset
SPAM messages          Good messages
1 missed                            1 missed
242 caught                         29 delivered
99.588% caught                3.333% missed
Total processed by filter
538 missed                         16 missed
4140 caught                      4434 delivered
From corpus 310 fed       8 fed

不过我的信头还是没有dspam 。是不是good的都不会经过dspam？
刚帮你排列了一下

[ 本帖最后由 liushaobo 于 2009-5-8 10:50 编辑 ]

回复引用

TOP

rodge

乞丐

4^#

发表于 2009-5-8 10:53 | 只看该作者

-----------------------------------一个客户的7天的  (并且有做白名单不过dspam)

Metric                                                                                  Calculated as
Overall accuracy (since last reset)       93.976%             (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset)    91.589%                (Spam catch rate only)
Spam ratio (of total processed)          48.846%                         Total SPAM messages (both caught & missed) / Total number of messages

                                                      SPAM messages             Good messages
Since last reset                      18 missed                         17 missed
                                             196 caught                      350 delivered
                                                91.589% caught          4.632% missed
Total processed by filter          395 missed             30 missed
                                             2822 caught                3339 delivered
From corpus                            235 fed                   11 fed

[ 本帖最后由 rodge 于 2009-5-8 10:58 编辑 ]

回复引用

TOP

archerhu

中级会员

Rank: 3 Rank: 3

5^#

发表于 2009-5-8 11:50 | 只看该作者

Performance Statistics - Fri May 8 11:49:50 2009
Metric Calculated as
Overall accuracy (since last reset) 99.593% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset) 99.569% (Spam catch rate only)
Spam ratio (of total processed) 89.791% Total SPAM messages (both caught & missed) / Total number of messages
SPAM messages Good messages
Since last reset 482 missed 6 missed
111260 caught 8114 delivered
99.569% caught 0.074% missed
Total processed by filter 1059 missed 82 missed
129389 caught 14750 delivered
From corpus 2392 fed 58 fed

开头效果很不错，这几天忽然差了很多，每天错判为正常邮件的垃圾邮件大概有个200封（400用户）

[ 本帖最后由 archerhu 于 2009-5-9 08:18 编辑 ]

回复引用

TOP

fzhyextmail

管理员

Rank: 9 Rank: 9 Rank: 9

Medal No.6

6^#

发表于 2009-5-11 16:04 | 只看该作者

原帖由 archerhu 于 2009-5-8 11:50 发表
Performance Statistics - Fri May 8 11:49:50 2009
Metric Calculated as
Overall accuracy (since last reset) 99.593% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam ide ...

贝叶斯的算法不是一次可以解决终身的，抓住近期明显的垃圾信件并强化正常邮件长期坚持训练，多看看贝叶斯算法描述

回复引用

TOP

hzqbbc

管理员

Rank: 9 Rank: 9 Rank: 9

Medal No.7 Medal No.10

7^#

发表于 2009-5-11 16:58 | 只看该作者

其实还需要结合smtp行为的过滤，才能相对完整点。我也发现dspam 过些时间，部分新的spam就会漏判.

请引用或摘抄本站文章信息的朋友，保留本站链接及作者信息，保护版权，谢谢。

构建高性能大容量开源邮件系统- ExtMail

Postfix在中国官方网站

回复引用

TOP

archerhu

中级会员

Rank: 3 Rank: 3

8^#

发表于 2009-5-12 23:44 | 只看该作者

原帖由 hzqbbc 于 2009-5-11 16:58 发表
其实还需要结合smtp行为的过滤，才能相对完整点。我也发现dspam 过些时间，部分新的spam就会漏判.

最近发现垃圾邮件报告不太好使，于是把正常邮件和垃圾邮件分开，用dspam_train训练，发现：
1.部分垃圾邮件无论学习多少次都学不会，可能因为内容为空，有附件的那些
2.其他垃圾邮件第一次学习一定不会pass，起码要两次，甚至大部分需要学习3-4次的才能pass。我用dspam命令对单个邮件样本学习也是必须第二次执行才能学会，但是对同类型的（内容几乎一样）的垃圾邮件还是判断不出来，由此开始严重怀疑我们平常报告垃圾邮件插件的效果

回复引用

TOP

chinaglwo

中级会员

Rank: 3 Rank: 3

9^#

发表于 2009-5-14 00:19 | 只看该作者

SPAM messages Good messages
Since last reset 0 missed 0 missed
13307 caught 44141 delivered
100.000% caught 0.000% missed
Total processed by filter 1967 missed 91 missed
60561 caught 178720 delivered
From corpus 4932 fed 150 fed

这是10天的记录

我多次进行了reset，但是邮件太多，手动retrain不过来了，正确率大概能在90%多点。

恩，我也发现有些邮件学习了多次后仍然学不会，以英文邮件居多

对中文垃圾邮件的处理，效果是非常好，直接进了“垃圾邮件”目录

回复引用

TOP

hzqbbc

管理员

Rank: 9 Rank: 9 Rank: 9

Medal No.7 Medal No.10

10^#

发表于 2009-5-14 07:45 | 只看该作者

看来还得结合行为过滤才行. 看看我们的

Metric Calculated as
Overall accuracy (since last reset) 97.869% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset) 98.025% (Spam catch rate only)
Spam ratio (of total processed) 65.028% Total SPAM messages (both caught & missed) / Total number of messages
SPAM messages Good messages
Since last reset 33 missed 3 missed
1638 caught 15 delivered
98.025% caught 16.667% missed
Total processed by filter 436 missed 20 missed
5200 caught 3011 delivered
From corpus 233 fed 14 fed

请引用或摘抄本站文章信息的朋友，保留本站链接及作者信息，保护版权，谢谢。

构建高性能大容量开源邮件系统- ExtMail

Postfix在中国官方网站

回复引用

TOP

ghost009

乞丐

11^#

发表于 2009-6-18 12:55 | 只看该作者

Metric Calculated as
Overall accuracy (since last reset) 93.836% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset) 87.205% (Spam catch rate only)
Spam ratio (of total processed) 43.462% Total SPAM messages (both caught & missed) / Total number of messages

SPAM messages Good messages
Since last reset 390 missed 42 missed
2658 caught 3919 delivered
87.205% caught 1.060% missed
Total processed by filter 390 missed 42 missed
2658 caught 3923 delivered
From corpus 215 fed 11 fed

回复引用

TOP

zhaodongxi

初级会员

Rank: 2

12^#

发表于 2009-9-25 15:49 | 只看该作者

总精准度 98.939% （捕捉成功的垃圾邮件＋已投递的正常邮件）÷ 邮件总数
垃圾邮件捕捉成功命中率 99.084% 捕捉成功的垃圾邮件 ÷ 垃圾邮件总数（包括已捕捉的和漏判的）
垃圾邮件比例 67.921% 垃圾邮件总数（包括已捕捉的和漏判的）÷ 邮件总数
垃圾邮件正常邮件
自上次重置统计后 92 漏判 48 漏判
9947 捕捉 3110 投递
99.084% 捕捉率 1.520% 漏判率
过滤统计 471 漏判 63 漏判
12580 捕捉 6101 投递
From corpus 251 fed 18 fed

[ 本帖最后由 zhaodongxi 于 2009-10-10 08:11 编辑 ]

回复引用

TOP

crastyl

新手上路

Rank: 1

13^#

发表于 2010-4-20 15:19 | 只看该作者

我的效果不大好
erformance Statistics - Tue Apr 20 15:06:57 2010
Metric Calculated as
Overall accuracy (since last reset) 65.797% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset) 50.638% (Spam catch rate only)
Spam ratio (of total processed) 50.116% Total SPAM messages (both caught & missed) / Total number of messages
SPAM messages Good messages
Since last reset 3365 missed 1248 missed
3452 caught 5422 delivered
50.638% caught 18.711% missed
Total processed by filter 3449 missed 1259 missed
3452 caught 5610 delivered
From corpus 9149 fed 335 fed

回复引用

TOP

crastyl

新手上路

Rank: 1

14^#

发表于 2010-4-20 15:19 | 只看该作者

erformance Statistics - Tue Apr 20 15:06:57 2010
Metric Calculated as
Overall accuracy (since last reset) 65.797% (SPAM messages caught + Good messages delivered) / Total number of messages
Spam identification (since last reset) 50.638% (Spam catch rate only)
Spam ratio (of total processed) 50.116% Total SPAM messages (both caught & missed) / Total number of messages
SPAM messages Good messages
Since last reset 3365 missed 1248 missed
3452 caught 5422 delivered
50.638% caught 18.711% missed
Total processed by filter 3449 missed 1259 missed
3452 caught 5610 delivered
From corpus 9149 fed 335 fed

回复引用

TOP

12 下一页

返回列表