Spam filters

Discussion in 'polls' started by JayK, Jun 6, 2003.

Thread Status:
Not open for further replies.
  1. JayK

    JayK Poster

    Joined:
    Dec 27, 2002
    Posts:
    619
    Just curious.

    I do 1-4
     
  2. Tinribs

    Tinribs Registered Member

    Joined:
    Mar 14, 2002
    Posts:
    734
    Location:
    England
    I do 1,2&3 but voted wrong :oops: K9 for me, bayesian filter,blacklist,whitelist and email client filter= 100% accuracy for over 6 weeks now
     
  3. meneer

    meneer Registered Member

    Joined:
    Nov 27, 2002
    Posts:
    1,132
    Location:
    The Netherlands
    Easy, I'm running spamassassin on my e-smith linux router.

    And one answer I'm missing: I love myrealbox.com, never in, say, 4 years have I received one spam mail. Great performance :D
     
  4. _Tat_

    _Tat_ Registered Member

    Joined:
    Apr 13, 2003
    Posts:
    14
    Location:
    somewhere between the worlds
    since recently most spamfiltering has been overtaken very efficiently by my email provider (gmx) - who is using a combination which can be described closest with 1-4. For my local spamfiltering a combination of 2-3 would meet the truth better, but that option was not given...
     
  5. JayK

    JayK Poster

    Joined:
    Dec 27, 2002
    Posts:
    619
    Sorry, there were just not enough options in the poll to cover everything and I didnt like a catchall , none of the above answer.

    My thoughts on the merits of each of the methods.

    Bayesian methods - Whether this is successful or not depends a lot on the proportion of spam you get. If you are using a bayesian filter that only classifies between spam and non-spam this is critical (e.g k9). If you have a very high porportion of spam (say over 70%), be ready to get a few false positives from first time correspondence from say websites you administer.

    If you are using a solution like Popfile, then the choice of folders (buckets according to PF terminology) you use is critical. Try to ensure each bucket gets at least 10% of mail classified. You can even split spam into 2 types..Adult and non-adult.

    Open relay black lists - my informal experiments show that they catch some but not all spam , inferior compared to Bayesian methods altough it depends a little on which black lists you use.

    Normal rules filtering on headers and body text can be effective (more so then bayesian methods that occasionally "act up" and let obvious spam through) at catching obvious spam, a numerical threshold like content control in the new Pegasus mail, Poccomail or Spamassasin can be effective.

    I havent tried the peer to peer methods.

    I put my trust in bayesian methods to catch some of the more subtle spam (e.g spam send through a yahoo mailing list i subscribe to) and back it up with SAproxy (mostly rely on their rulebase rather then their other methods)which usually catches everything else.


    Black and whitelists are reactive measures , I don't like them because they are too troublesome. Espically blacklists.
     
Thread Status:
Not open for further replies.