Re: [SLUG] {Spam?} Anti spam software?
2008/6/20 Sean Murphy [EMAIL PROTECTED]: All, I am after a good Anti spam software program for Linux which is shell based. I am aware of Spam assassign. But I would like to know if there is anything else which is better? I had amazingly good results using only bogofilter, after about a month training it on spam, and by hooking up my mail reader to also train it on non-spam (every message I replied to got pumped to bogofilter as non-spam training). My circumstantial evidence then is that bayesian filters work really really well when you train on both positive and negative inputs. I never got any non-spam marked as spam, and rarely did it misclassify spam. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] {Spam?} Anti spam software?
On Thu, Jun 19, 2008, Sean Murphy wrote: All, I am after a good Anti spam software program for Linux which is shell based. I am aware of Spam assassign. But I would like to know if there is anything else which is better? Better could mean a few things in the context of spam filtering, could you clarify which of these features is more important to you: 1. overall accuracy (false negatives and false positives) 2. fewest number of false negatives (spam that gets through to your inbox) 3. fewest number of false positives (good mail that ends up marked as spam) 4. good accuracy when in the default configuration, no twiddling required 5. good for processing large volumes of mail without insane resources I use SpamAssassin and find it does really well, but it falls down at #4 and #5. I have to train the Bayesian[1] classifiers on all my mail in order to get good-to-me accuracy, so I am certainly not relying on the default configuration. (I suspect I'd do just as well switching entirely to a Bayesian system, but since SpamAssassin is now doing fine I have not done so). And it's a resource hog, it sometimes takes 8 seconds to scan a mail on my OK-standard desktop system. So if you were receiving more than an email about every 8 seconds you'd be looking at performance tuning and additional less hoggy measures, or at alternatives. (Everything that processes the full body of an email is somewhat resource intensive, but I understand that SA is not great.) -Mary [1] Automatically trained and dynamic rules, rather than fixed rules. Has the downside that you need to fairly regularly show it current examples of good and bad mail so it can continue to learn to discriminate between them. The Bayesian machine learning techniques are not exactly cutting edge machine learning, but they only have to discriminate between two categories for this use. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] {Spam?} Anti spam software?
On Fri, 2008-06-20 at 07:57 +1000, Mary Gardiner wrote: On Thu, Jun 19, 2008, Sean Murphy wrote: All, I am after a good Anti spam software program for Linux which is shell based. I am aware of Spam assassign. But I would like to know if there is anything else which is better? Better could mean a few things in the context of spam filtering, could you clarify which of these features is more important to you: 1. overall accuracy (false negatives and false positives) 2. fewest number of false negatives (spam that gets through to your inbox) 3. fewest number of false positives (good mail that ends up marked as spam) 4. good accuracy when in the default configuration, no twiddling required 5. good for processing large volumes of mail without insane resources I use SpamAssassin and find it does really well, but it falls down at #4 and #5. I have to train the Bayesian[1] classifiers on all my mail in order to get good-to-me accuracy, so I am certainly not relying on the default configuration. (I suspect I'd do just as well switching entirely to a Bayesian system, but since SpamAssassin is now doing fine I have not done so). And it's a resource hog, it sometimes takes 8 seconds to scan a mail on my OK-standard desktop system. So if you were receiving more than an email about every 8 seconds you'd be looking at performance tuning and additional less hoggy measures, or at alternatives. (Everything that processes the full body of an email is somewhat resource intensive, but I understand that SA is not great.) I'm using Bogofilter (bayesian filter) to sort spam into good, bad or unsure at a user level. I've got a cron shell script that passes manually sorted unsure email through the filter hourly for training purposes and it works really well at the client level. I don't know if there is a package to do that. I wrote the script myself, which means that it is crude and simple :) This doesn't stop spam... it just means you never have to read it. David. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
Re: [SLUG] {Spam?} Anti spam software?
On Fri, 20 Jun 2008 12:50:20 +1000 david [EMAIL PROTECTED] wrote: On Fri, 2008-06-20 at 07:57 +1000, Mary Gardiner wrote: On Thu, Jun 19, 2008, Sean Murphy wrote: All, I am after a good Anti spam software program for Linux which is shell based. I am aware of Spam assassign. But I would like to know if there is anything else which is better? Better could mean a few things in the context of spam filtering, could you clarify which of these features is more important to you: 1. overall accuracy (false negatives and false positives) 2. fewest number of false negatives (spam that gets through to your inbox) 3. fewest number of false positives (good mail that ends up marked as spam) 4. good accuracy when in the default configuration, no twiddling required 5. good for processing large volumes of mail without insane resources I use SpamAssassin and find it does really well, but it falls down at #4 and #5. I have to train the Bayesian[1] classifiers on all my mail in order to get good-to-me accuracy, so I am certainly not relying on the default configuration. (I suspect I'd do just as well switching entirely to a Bayesian system, but since SpamAssassin is now doing fine I have not done so). And it's a resource hog, it sometimes takes 8 seconds to scan a mail on my OK-standard desktop system. So if you were receiving more than an email about every 8 seconds you'd be looking at performance tuning and additional less hoggy measures, or at alternatives. (Everything that processes the full body of an email is somewhat resource intensive, but I understand that SA is not great.) I'm using Bogofilter (bayesian filter) to sort spam into good, bad or unsure at a user level. I've got a cron shell script that passes manually sorted unsure email through the filter hourly for training purposes and it works really well at the client level. I don't know if there is a package to do that. I wrote the script myself, which means that it is crude and simple :) This doesn't stop spam... it just means you never have to read it. I use bogofilter in connection with Sylpheed as a simple pop client. It is remarkably effective, and with the default settings has almost no false positives. It is interesting with a Bayesian filter to notice that sometimes the spammers seem to try something new. A few spam messages get through, but with Sylpheed you just mark them as spam as they come in and it learns very quickly. Alan David. -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html -- Alan L Tyreehttp://www2.austlii.edu.au/~alan Tel: 04 2748 6206 Fax: +61 2 4782 7092 FWD: 615662 -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
