Junk Filter Wimpy? - Mozilla

This is a discussion on Junk Filter Wimpy? - Mozilla ; On 1/20/2008 3:33 AM, Ron K. wrote: > Wayne Mery keyboarded, On 1/20/2008 1:28 AM : >> On 1/14/2008 1:48 PM, Donald Cook wrote: >>> Due to nobody's fault other than my own, I suppose, I get to work on ...

+ Reply to Thread
Page 2 of 2 FirstFirst 1 2
Results 21 to 38 of 38

Thread: Junk Filter Wimpy?

  1. Re: Junk Filter Wimpy?

    On 1/20/2008 3:33 AM, Ron K. wrote:
    > Wayne Mery keyboarded, On 1/20/2008 1:28 AM :
    >> On 1/14/2008 1:48 PM, Donald Cook wrote:
    >>> Due to nobody's fault other than my own, I suppose, I get to work on
    >>> Monday morning with 500+ pieces of email, 90% of which are always
    >>> junk. Enlargers, drugs, Cyrillic announcements (sometimes in Russian,
    >>> sometimes transliterations of English ones, for whatever reason) or
    >>> Japanese ones, etc.
    >>>
    >>> This number is not going down, as I sit and individually mark the
    >>> junk ones as junk.
    >>>
    >>> Is there a way to get recognition going a bit more robustly? It takes
    >>> about a half hour on Monday, and fifteen minutes on other days.
    >>>
    >>> Thanks.
    >>>
    >>> Don.

    >>
    >> update 2.0.0.12 will deliver some JMC fixes.
    >> if it doesn't fix your issue then please file a bug.

    >
    > Any bug references for the patches ?
    >


    not offhand - please check rumbling edge and bugzilla for bugs that have
    keyword (I think) fixed1.8.1.12

  2. Re: Junk Filter Wimpy?

    goodwin wrote:
    > On 1/19/2008 11:25 PM Will Chapman scribbled:
    >
    >> Wayne
    >>
    >> Check out K9 from http://keir.net/k9.html I installed it 4 days
    >> ago and already it is picking up over 99% of my spam. I've still
    >> got TB junk switched on and it is interesting that TB doesn't
    >> pick up much of what K9 identifies.
    >>
    >> Cheers
    >>
    >>
    >> Will

    >
    > Tbird is probably trusting K9 tagging/whatever and not bothering to
    > check once K9 has passed on it. I don't know K9 but Tbird and Spampal
    > (a proxy) work that way. I recollect seeing somewhere (can't locate it
    > right now, may have been upon install of Spampal) a dialog asking if
    > Tbird should trust Spampal, which I answered yes to.


    TB only offers the ability to trust one or two specific external spam
    handlers, and K9 appears to be 'invisible' because it doesn't modify the
    headers that TB expects when using the "Trust xxx" setting.

    K9 does allow you to prepend/append text to the subject line so TB can
    learn to trust K9 using text filters, and K9 can be configured to add
    or modify xheaders, none of which TB knows how to identify as yet.

    The reason TB is missing the same messages K9 is getting is because
    that's the problem with TB's junk filter. Sorry.

  3. Re: Junk Filter Wimpy?

    Will Chapman wrote:
    > G. R. Woodring wrote:
    >> On 1/14/2008 6:41 PM, Donald Cook wrote:
    >>> Ron K. wrote:

    >
    >>>>
    >>>> The JAVA program is an inactive project at mozdev.org at this URL:
    >>>> http://bayesjunktool.mozdev.org/
    >>>>
    >>> Ron,
    >>>
    >>> Thanks for your suggestion.
    >>>
    >>> OK, looked at my training.dat file, and it is over 5MB. Larger than
    >>> you suggest keeping around. How can it get so big?
    >>>

    >
    > Had a look at mine and it is over 6mb and doesn't seem to be 'learning'
    > anymore.
    >
    > Before I switched to TB, I used to use K9 (http://keir.net) which
    > was 99% effective. I decided to download it again and try to set it up
    > with TB. It was a doddle. Just open up your Account properties, select
    > an account and:
    >
    > 1. Change port from 110 to 9999
    > 2. Change Pop server from, say, pop.my_own.net to 127.0.0.1
    > 3. Change account (user) name from, say My_Email@my_own.net,
    > to pop.my_own.net/110/My_Email@my_own.net.
    >
    > Repeat for any other spam-infested accounts.
    >
    > Open up K9 (there's an icon on your task tray) and wait (or invoke from
    > TB) for email to arrive. They will show on the K9
    > Recent Emails tab and, K9 will already have made an attempt to
    > guess what is Good Mail and Spam. Change those that are wrong.
    >
    > I installed it yesterday and it is already hitting in excess of 82%
    > accuracy. Coupled with TB built-in junk masher I'm catching
    > close to 100% of spam. Most of the work is being done by K9 because it
    > works by classifying spam before is leaves the server.
    >
    > I'm now going to reset the TB junk settings and see what happens.
    >
    > By the way, K9 is free.
    >
    > Cheers
    >
    > Will


    Although I've already posted this elsewhere, I'd like to second Will :
    K9 just works. Better than *any* Bayes filter *ever* did on this or two
    other systems. I got better results using K9 in the first week than I
    got over 3 years with 7 other Bayes tools. And that's using the EXACT
    SAME TRAINING DATA.

    And in the 2 months and 3 days since installing it, K9 has _completely
    eliminated_ my spam load. Completely. The best TB or any other Bayes
    filter did on the same data and incoming mail was around 45%, with up to
    25% false positives and 15-30% false negatives. I got tired of asking
    for help fixing the abysmal results with all the other Bayes tools,
    because I kept getting glad-handed with BS about "too much training" or
    "too little training" or "not allowing the filter to train by only
    manually training on false positives/false negatives". God, I heard it all.

    Out of the box, 3 minutes of configuring K9, then BAM instant accurate
    junk filtering, and it just gets better and better.

    So why does K9 work so well out of the box while ALL Bayes tools take at
    _least_ months to a year or more (in my case, never) to reach the same
    accuracy that K9 offers immediately? And why are so many users putting
    up with that kind of inefficiency with Bayes filters?

    Not to mention the sheer number of "what happened to my training data?
    It's disappeared and my Bayes filter has started out from scratch!!!"
    questions on this and all other Bayes forums I've watched.

    K9, on the other hand, with the same training data, and with less than
    0.5% of the training time, correctly identifies 99.981% of incoming
    email, with 0.059% false positives. No Bayes filter has ever achieved
    anywhere NEAR that kind of accuracy on this same system with the same
    data to work with.

    And K9 hasn't been updated in over two years! And it still works!

    I can't speak for the thousands of other good folks who use Bayes
    filters and it works for them just fine. In my case, none ever did. And
    I can't speak for all the users who like me suddenly lost their training
    sets, or had their dbs corrupted. Or whose Bayes filters just... stopped
    working. I had the same problems multiple times and I just assumed it
    was something I was doing wrong. But after 7 different tools, and
    probably 30 or more resets and reinstalls, I'm sticking to what actually
    works.

    If you have trouble with spam, and you can't get your TB junk filter or
    another Bayes filter reliably detecting false positives/negatives, or it
    keeps on missing the same old messages over and over again despite any
    training, and/or you've just lost a year's worth of training data, then
    at least try out K9. It's free (although if it works as well as it did
    on this system, I was happy to donate a little something to Robin Keir
    just for stopping the spam pain).

    So give Robin's little program a try - if TB's junk filter is as useless
    for you as it was for me, I think you'll be pleasantly surprised by K9.
    And I can just about guarantee that if you suffer from the same type of
    spam volume/load I do (52% spam per day, out of 39+ emails/day average)
    you'll notice an improvement from the minute you start using K9, and if
    you spend a few minutes (less than 10 minutes per week, max) maintaining
    and training the incredibly few false positives, you'll suddenly find
    that you can start to trust your inbox again. I did.

  4. Re: Junk Filter Wimpy?

    On 2/1/2008 5:20 PM PC Pete scribbled:

    > goodwin wrote:
    >> On 1/19/2008 11:25 PM Will Chapman scribbled:



    >>> I've still got TB junk switched on and it is interesting that TB doesn't
    >>> pick up much of what K9 identifies.



    >> I don't know K9 but Tbird and Spampal(a proxy) work that way. I recollect seeing somewhere (can't locate it
    >> right now, may have been upon install of Spampal) a dialog asking if
    >> Tbird should trust Spampal, which I answered yes to.

    >
    > TB only offers the ability to trust one or two specific external spam
    > handlers, and K9 appears to be 'invisible' because it doesn't modify the
    > headers that TB expects when using the "Trust xxx" setting.


    OK, I was speculating on why TB doesn't pick up what K9 does. My
    assumption is that once K9 picks up something, Tbird doesn't bother to
    look at it. I don't really know the mechanics involved with K9.

    >
    > K9 does allow you to prepend/append text to the subject line so TB can
    > learn to trust K9 using text filters, and K9 can be configured to add
    > or modify xheaders, none of which TB knows how to identify as yet.
    >
    > The reason TB is missing the same messages K9 is getting is because
    > that's the problem with TB's junk filter. Sorry.


    No reason to be sorry - I haven't an ax to grind here. I get your
    results using Spampal so who cares? Bayesian in and of itself won't
    cover spam 100% - not even 80% nowadays - IMO Bayesian filters are just
    about worthless to the average user.

  5. Re: Junk Filter Wimpy?

    goodwin wrote:
    > On 2/1/2008 5:20 PM PC Pete scribbled:
    >
    >> goodwin wrote:
    >>> On 1/19/2008 11:25 PM Will Chapman scribbled:

    >
    >
    >>>> I've still got TB junk switched on and it is interesting that TB doesn't
    >>>> pick up much of what K9 identifies.

    >
    >
    >>> I don't know K9 but Tbird and Spampal(a proxy) work that way. I recollect seeing somewhere (can't locate it
    >>> right now, may have been upon install of Spampal) a dialog asking if
    >>> Tbird should trust Spampal, which I answered yes to.

    >> TB only offers the ability to trust one or two specific external spam
    >> handlers, and K9 appears to be 'invisible' because it doesn't modify the
    >> headers that TB expects when using the "Trust xxx" setting.

    >
    > OK, I was speculating on why TB doesn't pick up what K9 does. My
    > assumption is that once K9 picks up something, Tbird doesn't bother to
    > look at it. I don't really know the mechanics involved with K9.
    >
    >> K9 does allow you to prepend/append text to the subject line so TB can
    >> learn to trust K9 using text filters, and K9 can be configured to add
    >> or modify xheaders, none of which TB knows how to identify as yet.
    >>
    >> The reason TB is missing the same messages K9 is getting is because
    >> that's the problem with TB's junk filter. Sorry.

    >
    > No reason to be sorry - I haven't an ax to grind here. I get your
    > results using Spampal so who cares? Bayesian in and of itself won't
    > cover spam 100% - not even 80% nowadays - IMO Bayesian filters are just
    > about worthless to the average user.



    Nonsense. My Thunderbird is currently running in excess of 99%
    reliability with regards to spam.

    i.e. If I see ONE spam in my inbox, there are 200 spam 'captured' in my
    junk folder. For every ONE spam I 'see', there are 200 (or more) I don't
    'see' (until I look in my Junk folder)

    And yes, my 'false positive' rate is currently at zero over the last six
    months - i.e. NO false positives since last spring.


    I don't call that 'about worthless' at all.


    And all I do is mark anything that is spam as 'junk' - and un-mark
    anything that isn't (tho I haven't done that in almost a year (un-mark
    that is))


  6. Re: Junk Filter Wimpy?

    PC Pete wrote:
    > Will Chapman wrote:
    >> G. R. Woodring wrote:
    >>> On 1/14/2008 6:41 PM, Donald Cook wrote:
    >>>> Ron K. wrote:

    >>
    >>>>>
    >>>>> The JAVA program is an inactive project at mozdev.org at this URL:
    >>>>> http://bayesjunktool.mozdev.org/
    >>>>>
    >>>> Ron,
    >>>>
    >>>> Thanks for your suggestion.
    >>>>
    >>>> OK, looked at my training.dat file, and it is over 5MB. Larger than
    >>>> you suggest keeping around. How can it get so big?
    >>>>

    >>
    >> Had a look at mine and it is over 6mb and doesn't seem to be
    >> 'learning' anymore.
    >>
    >> Before I switched to TB, I used to use K9 (http://keir.net) which
    >> was 99% effective. I decided to download it again and try to set it up
    >> with TB. It was a doddle. Just open up your Account properties, select
    >> an account and:
    >>
    >> 1. Change port from 110 to 9999
    >> 2. Change Pop server from, say, pop.my_own.net to 127.0.0.1
    >> 3. Change account (user) name from, say My_Email@my_own.net,
    >> to pop.my_own.net/110/My_Email@my_own.net.
    >>
    >> Repeat for any other spam-infested accounts.
    >>
    >> Open up K9 (there's an icon on your task tray) and wait (or invoke
    >> from TB) for email to arrive. They will show on the K9
    >> Recent Emails tab and, K9 will already have made an attempt to
    >> guess what is Good Mail and Spam. Change those that are wrong.
    >>
    >> I installed it yesterday and it is already hitting in excess of 82%
    >> accuracy. Coupled with TB built-in junk masher I'm catching
    >> close to 100% of spam. Most of the work is being done by K9 because it
    >> works by classifying spam before is leaves the server.
    >>
    >> I'm now going to reset the TB junk settings and see what happens.
    >>
    >> By the way, K9 is free.
    >>
    >> Cheers
    >>
    >> Will

    >
    > Although I've already posted this elsewhere, I'd like to second Will :
    > K9 just works. Better than *any* Bayes filter *ever* did on this or two
    > other systems. I got better results using K9 in the first week than I
    > got over 3 years with 7 other Bayes tools. And that's using the EXACT
    > SAME TRAINING DATA.
    >
    > And in the 2 months and 3 days since installing it, K9 has _completely
    > eliminated_ my spam load. Completely. The best TB or any other Bayes
    > filter did on the same data and incoming mail was around 45%, with up to
    > 25% false positives and 15-30% false negatives. I got tired of asking
    > for help fixing the abysmal results with all the other Bayes tools,
    > because I kept getting glad-handed with BS about "too much training" or
    > "too little training" or "not allowing the filter to train by only
    > manually training on false positives/false negatives". God, I heard it all.
    >
    > Out of the box, 3 minutes of configuring K9, then BAM instant accurate
    > junk filtering, and it just gets better and better.
    >
    > So why does K9 work so well out of the box while ALL Bayes tools take at
    > _least_ months to a year or more (in my case, never) to reach the same
    > accuracy that K9 offers immediately? And why are so many users putting
    > up with that kind of inefficiency with Bayes filters?
    >
    > Not to mention the sheer number of "what happened to my training data?
    > It's disappeared and my Bayes filter has started out from scratch!!!"
    > questions on this and all other Bayes forums I've watched.
    >
    > K9, on the other hand, with the same training data, and with less than
    > 0.5% of the training time, correctly identifies 99.981% of incoming
    > email, with 0.059% false positives. No Bayes filter has ever achieved
    > anywhere NEAR that kind of accuracy on this same system with the same
    > data to work with.
    >
    > And K9 hasn't been updated in over two years! And it still works!
    >
    > I can't speak for the thousands of other good folks who use Bayes
    > filters and it works for them just fine. In my case, none ever did. And
    > I can't speak for all the users who like me suddenly lost their training
    > sets, or had their dbs corrupted. Or whose Bayes filters just... stopped
    > working. I had the same problems multiple times and I just assumed it
    > was something I was doing wrong. But after 7 different tools, and
    > probably 30 or more resets and reinstalls, I'm sticking to what actually
    > works.
    >
    > If you have trouble with spam, and you can't get your TB junk filter or
    > another Bayes filter reliably detecting false positives/negatives, or it
    > keeps on missing the same old messages over and over again despite any
    > training, and/or you've just lost a year's worth of training data, then
    > at least try out K9. It's free (although if it works as well as it did
    > on this system, I was happy to donate a little something to Robin Keir
    > just for stopping the spam pain).
    >
    > So give Robin's little program a try - if TB's junk filter is as useless
    > for you as it was for me, I think you'll be pleasantly surprised by K9.
    > And I can just about guarantee that if you suffer from the same type of
    > spam volume/load I do (52% spam per day, out of 39+ emails/day average)
    > you'll notice an improvement from the minute you start using K9, and if
    > you spend a few minutes (less than 10 minutes per week, max) maintaining
    > and training the incredibly few false positives, you'll suddenly find
    > that you can start to trust your inbox again. I did.




    I have the same (or better) results as you simply by using the built in
    Thunderbird filter.
    As I mentioned in another post, my catch rate is 99% or better. For
    every ONE spam I see in my inbox, there are 200 (or more) spams filtered
    out by Thunderbird.

    I haven't had a false positive in over six months.

    I don't call that worthless. I call that pretty durn good in fact.

    Time spent training? Only as long as it takes to press the JUNK button
    (and the NOT JUNK button after reviewing the catch*)

    *used to be, havent had to do that since last spring of course

  7. Re: Junk Filter Wimpy?

    On 2/2/2008 1:43 AM Moz Champion (Dan) scribbled:

    > goodwin wrote:
    >> On 2/1/2008 5:20 PM PC Pete scribbled:
    >>
    >>> goodwin wrote:
    >>>> On 1/19/2008 11:25 PM Will Chapman scribbled:

    >>
    >>>>> I've still got TB junk switched on and it is interesting that TB doesn't
    >>>>> pick up much of what K9 identifies.

    >>
    >>>> I don't know K9 but Tbird and Spampal(a proxy) work that way. I recollect seeing somewhere (can't locate it
    >>>> right now, may have been upon install of Spampal) a dialog asking if
    >>>> Tbird should trust Spampal, which I answered yes to.
    >>> TB only offers the ability to trust one or two specific external spam
    >>> handlers, and K9 appears to be 'invisible' because it doesn't modify the
    >>> headers that TB expects when using the "Trust xxx" setting.

    >> OK, I was speculating on why TB doesn't pick up what K9 does. My
    >> assumption is that once K9 picks up something, Tbird doesn't bother to
    >> look at it. I don't really know the mechanics involved with K9.
    >>
    >>> K9 does allow you to prepend/append text to the subject line so TB can
    >>> learn to trust K9 using text filters, and K9 can be configured to add
    >>> or modify xheaders, none of which TB knows how to identify as yet.
    >>>
    >>> The reason TB is missing the same messages K9 is getting is because
    >>> that's the problem with TB's junk filter. Sorry.

    >> No reason to be sorry - I haven't an ax to grind here. I get your
    >> results using Spampal so who cares? Bayesian in and of itself won't
    >> cover spam 100% - not even 80% nowadays - IMO Bayesian filters are just
    >> about worthless to the average user.

    >
    >
    > Nonsense. My Thunderbird is currently running in excess of 99%
    > reliability with regards to spam.


    been down this road before...

    >
    > i.e. If I see ONE spam in my inbox, there are 200 spam 'captured' in my
    > junk folder. For every ONE spam I 'see', there are 200 (or more) I don't
    > 'see' (until I look in my Junk folder)
    >
    > And yes, my 'false positive' rate is currently at zero over the last six
    > months - i.e. NO false positives since last spring.
    >
    >
    > I don't call that 'about worthless' at all.


    spammers can evade bayesian filters easier than just about any others.
    Worthless was an unfortunate choice of words on my part, however.
    Between my proxy and my ISPs filters, I get better results than with tb.
    Of course, YMMV.

    >
    >
    > And all I do is mark anything that is spam as 'junk' - and un-mark
    > anything that isn't (tho I haven't done that in almost a year (un-mark
    > that is))
    >


  8. Re: Junk Filter Wimpy?

    On 2/1/2008 8:37 PM, PC Pete wrote:

    > Will Chapman wrote:
    >> G. R. Woodring wrote:
    >>> On 1/14/2008 6:41 PM, Donald Cook wrote:
    >>>> Ron K. wrote:
    >>>>> The JAVA program is an inactive project at mozdev.org at this
    >>>>> URL: http://bayesjunktool.mozdev.org/
    >>>>>
    >>>> Ron,
    >>>>
    >>>> Thanks for your suggestion.
    >>>>
    >>>> OK, looked at my training.dat file, and it is over 5MB. Larger
    >>>> than you suggest keeping around. How can it get so big?
    >>>>

    >> Had a look at mine and it is over 6mb and doesn't seem to be
    >> 'learning' anymore.
    >>
    >> Before I switched to TB, I used to use K9 (http://keir.net) which
    >> was 99% effective. I decided to download it again and try to set it
    >> up with TB. It was a doddle. Just open up your Account properties,
    >> select an account and:
    >>
    >> 1. Change port from 110 to 9999 2. Change Pop server from, say,
    >> pop.my_own.net to 127.0.0.1 3. Change account (user) name from, say
    >> My_Email@my_own.net, to pop.my_own.net/110/My_Email@my_own.net.
    >>
    >> Repeat for any other spam-infested accounts.
    >>
    >> Open up K9 (there's an icon on your task tray) and wait (or invoke
    >> from TB) for email to arrive. They will show on the K9 Recent
    >> Emails tab and, K9 will already have made an attempt to guess what
    >> is Good Mail and Spam. Change those that are wrong.
    >>
    >> I installed it yesterday and it is already hitting in excess of 82%
    >> accuracy. Coupled with TB built-in junk masher I'm catching close
    >> to 100% of spam. Most of the work is being done by K9 because it
    >> works by classifying spam before is leaves the server.
    >>
    >> I'm now going to reset the TB junk settings and see what happens.
    >>
    >> By the way, K9 is free.
    >>
    >> Cheers
    >>
    >> Will

    >
    > Although I've already posted this elsewhere, I'd like to second Will
    > : K9 just works. Better than *any* Bayes filter *ever* did on this or
    > two other systems. I got better results using K9 in the first week
    > than I got over 3 years with 7 other Bayes tools. And that's using
    > the EXACT SAME TRAINING DATA.
    >
    > And in the 2 months and 3 days since installing it, K9 has
    > _completely eliminated_ my spam load. Completely. The best TB or any
    > other Bayes filter did on the same data and incoming mail was around
    > 45%, with up to 25% false positives and 15-30% false negatives. I got
    > tired of asking for help fixing the abysmal results with all the
    > other Bayes tools, because I kept getting glad-handed with BS about
    > "too much training" or "too little training" or "not allowing the
    > filter to train by only manually training on false positives/false
    > negatives". God, I heard it all.
    >
    > Out of the box, 3 minutes of configuring K9, then BAM instant
    > accurate junk filtering, and it just gets better and better.
    >
    > So why does K9 work so well out of the box while ALL Bayes tools take
    > at _least_ months to a year or more (in my case, never) to reach the
    > same accuracy that K9 offers immediately? And why are so many users
    > putting up with that kind of inefficiency with Bayes filters?
    >
    > Not to mention the sheer number of "what happened to my training
    > data? It's disappeared and my Bayes filter has started out from
    > scratch!!!" questions on this and all other Bayes forums I've
    > watched.
    >
    > K9, on the other hand, with the same training data, and with less
    > than 0.5% of the training time, correctly identifies 99.981% of
    > incoming email, with 0.059% false positives. No Bayes filter has ever
    > achieved anywhere NEAR that kind of accuracy on this same system with
    > the same data to work with.
    >
    > And K9 hasn't been updated in over two years! And it still works!
    >
    > I can't speak for the thousands of other good folks who use Bayes
    > filters and it works for them just fine. In my case, none ever did.
    > And I can't speak for all the users who like me suddenly lost their
    > training sets, or had their dbs corrupted. Or whose Bayes filters
    > just... stopped working. I had the same problems multiple times and I
    > just assumed it was something I was doing wrong. But after 7
    > different tools, and probably 30 or more resets and reinstalls, I'm
    > sticking to what actually works.
    >
    > If you have trouble with spam, and you can't get your TB junk filter
    > or another Bayes filter reliably detecting false positives/negatives,
    > or it keeps on missing the same old messages over and over again
    > despite any training, and/or you've just lost a year's worth of
    > training data, then at least try out K9. It's free (although if it
    > works as well as it did on this system, I was happy to donate a
    > little something to Robin Keir just for stopping the spam pain).
    >
    > So give Robin's little program a try - if TB's junk filter is as
    > useless for you as it was for me, I think you'll be pleasantly
    > surprised by K9. And I can just about guarantee that if you suffer
    > from the same type of spam volume/load I do (52% spam per day, out of
    > 39+ emails/day average) you'll notice an improvement from the minute
    > you start using K9, and if you spend a few minutes (less than 10
    > minutes per week, max) maintaining and training the incredibly few
    > false positives, you'll suddenly find that you can start to trust
    > your inbox again. I did.


    I've had the same problems as you used to and I get an average of 250
    emails a day. Thunderbird has never had better than an 80% or so
    success rate, and I have done all the things you mentioned - trained,
    marked, unmarked, etc - through many versions and it has never changed
    for the better. The problem for me has always been false negatives;
    false positives have always worked relatively well. So, for the hell of
    it, and after reading your post, I decided to try K9. That was
    yesterday. After less than one day and 196 emails at this writing, I am
    at 95.41% and getting better.

    Many thanks for your post.

    --
    Best regards
    Gord McFee

  9. Re: Junk Filter Wimpy?

    Moz Champion (Dan) wrote:
    > goodwin wrote:
    >> On 2/1/2008 5:20 PM PC Pete scribbled:
    >>
    >>> goodwin wrote:
    >>>> On 1/19/2008 11:25 PM Will Chapman scribbled:

    >>
    >>
    >>>>> I've still got TB junk switched on and it is interesting that TB
    >>>>> doesn't
    >>>>> pick up much of what K9 identifies.

    >>
    >>
    >>>> I don't know K9 but Tbird and Spampal(a proxy) work that way. I
    >>>> recollect seeing somewhere (can't locate it
    >>>> right now, may have been upon install of Spampal) a dialog asking if
    >>>> Tbird should trust Spampal, which I answered yes to.
    >>> TB only offers the ability to trust one or two specific external spam
    >>> handlers, and K9 appears to be 'invisible' because it doesn't modify
    >>> the headers that TB expects when using the "Trust xxx" setting.

    >>
    >> OK, I was speculating on why TB doesn't pick up what K9 does. My
    >> assumption is that once K9 picks up something, Tbird doesn't bother to
    >> look at it. I don't really know the mechanics involved with K9.
    >>
    >>> K9 does allow you to prepend/append text to the subject line so TB
    >>> can learn to trust K9 using text filters, and K9 can be configured
    >>> to add or modify xheaders, none of which TB knows how to identify as
    >>> yet.
    >>>
    >>> The reason TB is missing the same messages K9 is getting is because
    >>> that's the problem with TB's junk filter. Sorry.

    >>
    >> No reason to be sorry - I haven't an ax to grind here. I get your
    >> results using Spampal so who cares? Bayesian in and of itself won't
    >> cover spam 100% - not even 80% nowadays - IMO Bayesian filters are just
    >> about worthless to the average user.

    >
    >
    > Nonsense. My Thunderbird is currently running in excess of 99%
    > reliability with regards to spam.
    >
    > i.e. If I see ONE spam in my inbox, there are 200 spam 'captured' in my
    > junk folder. For every ONE spam I 'see', there are 200 (or more) I don't
    > 'see' (until I look in my Junk folder)
    >
    > And yes, my 'false positive' rate is currently at zero over the last six
    > months - i.e. NO false positives since last spring.
    >
    >
    > I don't call that 'about worthless' at all.
    >
    >
    > And all I do is mark anything that is spam as 'junk' - and un-mark
    > anything that isn't (tho I haven't done that in almost a year (un-mark
    > that is))
    >

    I am having the same experience, using plane old garden variety TB!!
    Unbelievable, almost!! I have had only one false positive in the past 6
    months, only two carefully engineered spams slipped in. Have dumped
    between 250 and 300 out of my junk file each week! As far as I am
    concerned there is no excuse for cluttering up a machine with anything
    other than a well trained Thunderbird!!

    Big John

  10. Re: Junk Filter Wimpy?

    Moz Champion (Dan) wrote:
    > goodwin wrote:
    >> On 2/1/2008 5:20 PM PC Pete scribbled:
    >>
    >>> goodwin wrote:
    >>>> On 1/19/2008 11:25 PM Will Chapman scribbled:

    >>
    >>
    >>>>> I've still got TB junk switched on and it is interesting that TB
    >>>>> doesn't
    >>>>> pick up much of what K9 identifies.

    >>
    >>
    >>>> I don't know K9 but Tbird and Spampal(a proxy) work that way. I
    >>>> recollect seeing somewhere (can't locate it
    >>>> right now, may have been upon install of Spampal) a dialog asking if
    >>>> Tbird should trust Spampal, which I answered yes to.
    >>> TB only offers the ability to trust one or two specific external spam
    >>> handlers, and K9 appears to be 'invisible' because it doesn't modify
    >>> the headers that TB expects when using the "Trust xxx" setting.

    >>
    >> OK, I was speculating on why TB doesn't pick up what K9 does. My
    >> assumption is that once K9 picks up something, Tbird doesn't bother to
    >> look at it. I don't really know the mechanics involved with K9.


    I don't think that is the case in the way that I have my system
    set up. I have tried switching TB junk system off and the spam
    immediately slips through again and despite training each day for
    several years (and re-building the training file) I have never
    achieved better than 75% with TB.

    I switch k9 back on again and leave the spam it identifies
    unmarked and TB catches about 50% of it. After just a few hours
    with k9 that was exceeded, after a few of days I was up to 90%
    and after 2 weeks I am 100%.

    > Nonsense. My Thunderbird is currently running in excess of 99%
    > reliability with regards to spam.
    >


    It is interesting how people's experience with TB is so widely
    different. I guess it is the nature of spam being received.

    So the message is clear, in the eventuality that isn't working
    well enough for anyone, I'm just suggesting that you try K9.

    Cheers

    Will

  11. Re: Junk Filter Wimpy?

    On 2/3/2008 5:01 AM, Will Chapman wrote:

    > Moz Champion (Dan) wrote:
    >> goodwin wrote:
    >>> On 2/1/2008 5:20 PM PC Pete scribbled:
    >>>
    >>>> goodwin wrote:
    >>>>> On 1/19/2008 11:25 PM Will Chapman scribbled:
    >>>
    >>>>>> I've still got TB junk switched on and it is interesting
    >>>>>> that TB doesn't pick up much of what K9 identifies.
    >>>
    >>>>> I don't know K9 but Tbird and Spampal(a proxy) work that way.
    >>>>> I recollect seeing somewhere (can't locate it right now, may
    >>>>> have been upon install of Spampal) a dialog asking if Tbird
    >>>>> should trust Spampal, which I answered yes to.
    >>>> TB only offers the ability to trust one or two specific
    >>>> external spam handlers, and K9 appears to be 'invisible'
    >>>> because it doesn't modify the headers that TB expects when
    >>>> using the "Trust xxx" setting.
    >>> OK, I was speculating on why TB doesn't pick up what K9 does. My
    >>> assumption is that once K9 picks up something, Tbird doesn't
    >>> bother to look at it. I don't really know the mechanics involved
    >>> with K9.

    >
    > I don't think that is the case in the way that I have my system set
    > up. I have tried switching TB junk system off and the spam
    > immediately slips through again and despite training each day for
    > several years (and re-building the training file) I have never
    > achieved better than 75% with TB.
    >
    > I switch k9 back on again and leave the spam it identifies unmarked
    > and TB catches about 50% of it. After just a few hours with k9 that
    > was exceeded, after a few of days I was up to 90% and after 2 weeks I
    > am 100%.
    >
    >> Nonsense. My Thunderbird is currently running in excess of 99%
    >> reliability with regards to spam.
    >>

    >
    > It is interesting how people's experience with TB is so widely
    > different. I guess it is the nature of spam being received.
    >
    > So the message is clear, in the eventuality that isn't working well
    > enough for anyone, I'm just suggesting that you try K9.


    You may have seen my post from yesterday. At that time, my stats after
    less than one day with K9, and 196 emails at that time, were 95.41%.
    Now, after less than two days and 439 emails in (I am averaging about
    257 a day), I am at 97.95%. This is truly amazing. I should mention
    that I post to the Internet under my real email address and always have
    and, as noted, receive a high volume of email each day. It may be
    partially these two facts that cause Thunderbird not to do better than
    80% for me. Interestingly, Forte Agent's spam feature caught 98.5% on
    386 emails the last time I used it. I suppose if I posted to the
    Internet under an alias, and only received 35 to 50 emails a day,
    Thunderbird alone would do the trick, but in my situation at least, K9
    is a terrific find.

    --
    Best regards
    Gord McFee

  12. Re: Junk Filter Wimpy?

    On 2/3/2008 1:06 PM Gord McFee scribbled:



    >
    > You may have seen my post from yesterday. At that time, my stats after
    > less than one day with K9, and 196 emails at that time, were 95.41%.
    > Now, after less than two days and 439 emails in (I am averaging about
    > 257 a day), I am at 97.95%. This is truly amazing. I should mention
    > that I post to the Internet under my real email address and always have
    > and, as noted, receive a high volume of email each day. It may be
    > partially these two facts that cause Thunderbird not to do better than
    > 80% for me.


    I doubt this affects Tbird and its filters. When you say post to the
    Internet, do you mean newsgroups via NNTP?
    If so, its been shown that newsgroups don't really get harvested for
    addresses - /web sites/ do.


    Interestingly, Forte Agent's spam feature caught 98.5% on
    > 386 emails the last time I used it.


    Agent uses regex more than other stuff, doesn't it?

    I suppose if I posted to the
    > Internet under an alias, and only received 35 to 50 emails a day,
    > Thunderbird alone would do the trick, but in my situation at least, K9
    > is a terrific find.
    >


    Does K9 use dnsbls?
    Intuitively, I think Tbird works better as the volume goes up. More
    likely, the spam you are getting is making a conscious effort to thwart
    baysian filters. Some spammers do that a lot more than others, mostly
    the pillz sellers. For me, I'm getting mostly 419 phishes for the past
    month after having not seen any for years. Spampal regex picks most of
    these off by simple word matching, doubt Tbird even looks at it since it
    trusts Spampal.

  13. OT Re: Junk Filter Wimpy?

    On 2/3/2008 2:19 PM goodwin scribbled:

    > On 2/3/2008 1:06 PM Gord McFee scribbled:
    >
    >
    >
    >> You may have seen my post from yesterday. At that time, my stats after
    >> less than one day with K9, and 196 emails at that time, were 95.41%.
    >> Now, after less than two days and 439 emails in (I am averaging about
    >> 257 a day), I am at 97.95%. This is truly amazing. I should mention
    >> that I post to the Internet under my real email address and always have
    >> and, as noted, receive a high volume of email each day. It may be
    >> partially these two facts that cause Thunderbird not to do better than
    >> 80% for me.

    >
    > I doubt this affects Tbird and its filters. When you say post to the
    > Internet, do you mean newsgroups via NNTP?
    > If so, its been shown that newsgroups don't really get harvested for
    > addresses - /web sites/ do.
    >
    >
    > Interestingly, Forte Agent's spam feature caught 98.5% on
    >> 386 emails the last time I used it.

    >
    > Agent uses regex more than other stuff, doesn't it?
    >
    > I suppose if I posted to the
    >> Internet under an alias, and only received 35 to 50 emails a day,
    >> Thunderbird alone would do the trick, but in my situation at least, K9
    >> is a terrific find.
    >>

    >
    > Does K9 use dnsbls?
    > Intuitively, I think Tbird works better as the volume goes up. More
    > likely, the spam you are getting is making a conscious effort to thwart
    > baysian filters. Some spammers do that a lot more than others, mostly
    > the pillz sellers. For me, I'm getting mostly 419 phishes for the past
    > month after having not seen any for years. Spampal regex picks most of
    > these off by simple word matching, doubt Tbird even looks at it since it
    > trusts Spampal.


    Here is a spam that no one picked up - neither my ISP, Tbird, nor Spampal.
    And its not in English...
    http://www.spamcop.net/sc?id=z163708...24e5101fde347z

  14. Re: Junk Filter Wimpy?

    On 2/3/2008 5:19 PM, goodwin wrote:

    > On 2/3/2008 1:06 PM Gord McFee scribbled:
    >
    >
    >
    >> You may have seen my post from yesterday. At that time, my stats
    >> after less than one day with K9, and 196 emails at that time, were
    >> 95.41%. Now, after less than two days and 439 emails in (I am
    >> averaging about 257 a day), I am at 97.95%. This is truly amazing.
    >> I should mention that I post to the Internet under my real email
    >> address and always have and, as noted, receive a high volume of
    >> email each day. It may be partially these two facts that cause
    >> Thunderbird not to do better than 80% for me.

    >
    > I doubt this affects Tbird and its filters. When you say post to the
    > Internet, do you mean newsgroups via NNTP? If so, its been shown
    > that newsgroups don't really get harvested for addresses - /web
    > sites/ do.


    I guess I do too many of both. :-)

    > Interestingly, Forte Agent's spam feature caught 98.5% on
    >> 386 emails the last time I used it.

    >
    > Agent uses regex more than other stuff, doesn't it?


    Don't know the internals. Maybe Ralph Fox will see this.

    > I suppose if I posted to the
    >> Internet under an alias, and only received 35 to 50 emails a day,
    >> Thunderbird alone would do the trick, but in my situation at least,
    >> K9 is a terrific find.

    >
    > Does K9 use dnsbls? Intuitively, I think Tbird works better as the
    > volume goes up. More likely, the spam you are getting is making a
    > conscious effort to thwart baysian filters. Some spammers do that a
    > lot more than others, mostly the pillz sellers. For me, I'm getting
    > mostly 419 phishes for the past month after having not seen any for
    > years. Spampal regex picks most of these off by simple word
    > matching, doubt Tbird even looks at it since it trusts Spampal.


    K9 just seems to be a better filter, I'm not sure why, but it works. By
    the way, now at 98.19 on 496 emails.

    --
    Best regards
    Gord McFee

  15. Re: Junk Filter Wimpy?

    goodwin wrote:
    > On 2/3/2008 1:06 PM Gord McFee scribbled:
    >
    >> Now, after less than two days and 439 emails in (I am averaging about
    >> 257 a day), I am at 97.95%. This is truly amazing. I should mention
    >> that I post to the Internet under my real email address and always have
    >> and, as noted, receive a high volume of email each day. It may be
    >> partially these two facts that cause Thunderbird not to do better than
    >> 80% for me.

    >
    > I doubt this affects Tbird and its filters. When you say post to the
    > Internet, do you mean newsgroups via NNTP?
    > If so, its been shown that newsgroups don't really get harvested for
    > addresses - /web sites/ do.
    >


    Well, I have three email addresses that I use reguarly. One is
    on a website used to be undisguised and I get about 30 posts a
    day on that and the spam rate is over 80% which matches your
    experience.

    Then I have one that is used on newsgroups and I get around
    175-200 emails a day on that which is 80% spam. So that doesn't
    match your experience.

    The third is also used regularly but in less volume for business
    mail - I get about 20 posts a day and about 75% is spam; the fact
    that I restrict it only for business use suggests darker
    ativities at play.
    >
    > Interestingly, Forte Agent's spam feature caught 98.5% on
    >> 386 emails the last time I used it.

    >
    > Agent uses regex more than other stuff, doesn't it?
    >
    > I suppose if I posted to the
    >> Internet under an alias, and only received 35 to 50 emails a day,
    >> Thunderbird alone would do the trick, but in my situation at least, K9
    >> is a terrific find.
    >>

    >
    > Does K9 use dnsbls?


    If you select to do so.

    Cheers


    Will

  16. Re: Junk Filter Wimpy?

    On 2/4/2008 4:08 PM Will Chapman scribbled:

    > goodwin wrote:
    >> On 2/3/2008 1:06 PM Gord McFee scribbled:
    >>
    >>> Now, after less than two days and 439 emails in (I am averaging about
    >>> 257 a day), I am at 97.95%. This is truly amazing. I should mention
    >>> that I post to the Internet under my real email address and always have
    >>> and, as noted, receive a high volume of email each day. It may be
    >>> partially these two facts that cause Thunderbird not to do better than
    >>> 80% for me.

    >> I doubt this affects Tbird and its filters. When you say post to the
    >> Internet, do you mean newsgroups via NNTP?
    >> If so, its been shown that newsgroups don't really get harvested for
    >> addresses - /web sites/ do.
    >>

    >
    > Well, I have three email addresses that I use reguarly. One is
    > on a website used to be undisguised and I get about 30 posts a
    > day on that and the spam rate is over 80% which matches your
    > experience.
    >
    > Then I have one that is used on newsgroups and I get around
    > 175-200 emails a day on that which is 80% spam. So that doesn't
    > match your experience.


    My "experience" is that after masking my addy for years on newsgroups,
    I decided to use my regular address. I did this after reading in
    several antispam groups that addresses just don't get scraped. All
    those posters used real addresses and most are sys admins. When I
    switched, my spam load didn't change, even went down a tad (but thats
    due to my ISP getting more aggressive with their filters).

    >
    > The third is also used regularly but in less volume for business
    > mail - I get about 20 posts a day and about 75% is spam; the fact
    > that I restrict it only for business use suggests darker
    > ativities at play.


    Dunno what darker forces may be but my wife set up an address just for
    business and was being spammed within a week w/o using it.
    She says cox sold it, I doubt that and I have no love lost for cox...

  17. Re: Junk Filter Wimpy?

    But K9 doesn't work with SSL, sadly. ..........Edgar

    Will Chapman wrote:
    > Wayne Mery wrote:
    >> On 1/15/2008 6:11 PM, Rich Pasco wrote:
    >>> goodwin wrote:
    >>>
    >>>> On 1/14/2008 3:41 PM Donald Cook scribbled:
    >>>>> So the only option now is to get rid of my swollen training.dat file?
    >>>> Won't hurt to nuke it.
    >>> Is there a difference in effect between deleting the file training.dat
    >>> and invoking Tools / Options / Privacy / Junk / Reset Training Data ?
    >>> If so, what is it?
    >>>
    >>> - Rich

    >
    > Wayne
    >
    > Check out K9 from http://keir.net/k9.html I installed it 4 days
    > ago and already it is picking up over 99% of my spam. I've still
    > got TB junk switched on and it is interesting that TB doesn't
    > pick up much of what K9 identifies.
    >
    > Cheers
    >
    >
    > Will
    > _______________________________________________
    > support-thunderbird mailing list
    > support-thunderbird@lists.mozilla.org
    > https://lists.mozilla.org/listinfo/support-thunderbird
    > To unsubscribe, send an email to support-thunderbird-request@lists.mozilla.org?subject=unsubscribe
    >


    --
    Edgar G. Goodaire Phone: 709-737-8097
    Dept. of Mathematics and Statistics Fax: 709-737-3010
    Memorial University email: edgar@math.mun.ca
    St. John's, Newfoundland http://www.math.mun.ca/~edgar
    Canada A1C 5S7

  18. Re: Junk Filter Wimpy?

    I have been using POPFILE for a few years now... with a 99.6% success
    rate... recently moved to TB and POPFILE is working perfectly.. highly
    recommended!!!

    http://popfile.sourceforge.net/


    Edgar Goodaire wrote:
    > But K9 doesn't work with SSL, sadly. ..........Edgar
    >
    > Will Chapman wrote:
    >> Wayne Mery wrote:
    >>> On 1/15/2008 6:11 PM, Rich Pasco wrote:
    >>>> goodwin wrote:
    >>>>
    >>>>> On 1/14/2008 3:41 PM Donald Cook scribbled:
    >>>>>> So the only option now is to get rid of my swollen training.dat file?
    >>>>> Won't hurt to nuke it.
    >>>> Is there a difference in effect between deleting the file training.dat
    >>>> and invoking Tools / Options / Privacy / Junk / Reset Training Data ?
    >>>> If so, what is it?
    >>>>
    >>>> - Rich

    >>
    >> Wayne
    >>
    >> Check out K9 from http://keir.net/k9.html I installed it 4 days ago
    >> and already it is picking up over 99% of my spam. I've still
    >> got TB junk switched on and it is interesting that TB doesn't
    >> pick up much of what K9 identifies.
    >>
    >> Cheers
    >>
    >>
    >> Will
    >> _______________________________________________
    >> support-thunderbird mailing list
    >> support-thunderbird@lists.mozilla.org
    >> https://lists.mozilla.org/listinfo/support-thunderbird
    >> To unsubscribe, send an email to
    >> support-thunderbird-request@lists.mozilla.org?subject=unsubscribe
    >>

    >


+ Reply to Thread
Page 2 of 2 FirstFirst 1 2