Roady said:
Yeah, and why not install Linux if you don't want to wait for Windows
Vista?
C'mon Vanguard!
The corporate thing you mention is not the reason why there is no "Train"
button. The reason is more; Why would you want to anyway? Don't you want a
filter that just works out of the box and not after weeks and weeks of
fiddling with it and feeding it spam samples? I waste enough time on spam
already so I really want to spent anymore time on it by configuring it or
saving its settings so I can also get itworking on my laptop. Also; what
if
I train it wrong? Things are not so different for home or corporate users
here. As a matter of fact the Junk E-mail Filter is more a "home user"
feature anyway as in corporate environments filtering is more likely done
at
firewall or mailserver level. Also the (continously developed) filter that
we see in Outlook will feature in Windows Mail on Windows Vista.
As for me; even on the lowest setting the filter catches way over 99% of
my
spam (to test this I'm using a mailbox that only receives spam which is
currently around 250 messages a day. I let it pile up for some time,
create
a new mail profile and download all the messages at once and see what it
catches. On a rainy day I even compare the results I get from running it
with "out of the box" Outlook, the previous junk e-mail update filter and
the latest)
So you are claiming the OP should suffer with OL2003 *not* performing well
for him just because it performs well for you? I didn't say to get rid of
Outlook (as you imply with your irrelevant Vista statement). I said to
augment Outlook - or whatever e-mail client - with better spam filtering
software (and which you yourself mention is done in a corporate
environment). If the OP wants *better* spam filtering than what is in
OL2003, they should *not* wait around for some ambiguous and unlikely event
that will only occur sometime in the future if they want to get rid of the
spam now.
Just how did you extrapolate my "corporate thing" to be an explanation for
the absence of a train function for the Bayesian filter? The "corporate
thing" was in regards as to how well (or poorly) Microsoft will respond to
*end-user* requests versus feedback from their corporate customers.
A preset database of word weighting based on someone else's experience will
not reflect the content of e-mails that YOU have receive (and Bayesian
filtering is all based on *history*). That is the point of training a
Bayesian filter based on a history of good and bad e-mails that YOU have
received. It reduces the number of false positives because the weightings,
if even present, are based on your personal history of e-mails and not on
someone else's. Bayesian filtering is still a guessing algorithm and that
means it WILL have false positives. A false positive is more critical than
some false negatives. Does OL2003 even provide configurability of a "noise"
option to expire old words from its database that are no longer present in
recent e-mails to eliminate the poisoning attempted by spammers with their
garbage content in their spam? I suppose Microsoft could've hardcoded in
some expiry of old words but then it would not represent a value based on
the volume of e-mails experienced by a particular user and the user doesn't
get to decide on how much of a history to retain in the database.
It's nice that OL2003's spam filtering works so great for you. I am curious
to know how OL2003 detects any spam that uses the .gif attachment to hide
the content of the spam message. That is, does OL2003 drill into the
attachments to also interrogate their contents and include them in
weighting? If so, does the user get to specify WHICH filetypes are to be
included or excluded from that interrogation, so .gif files would be
included but .doc files would not? Didn't think so because it can't
interrogate a graphics file for text content. That means all that spam that
uses a .gif payload to present the spam will not get detected by OL2003's
spam filter. Far more than 1% of the spam that I receive still uses the
..gif payload to hide its content from Bayesian or other word-search filters.
You need to employ other schemes to detect spam than just Bayesian filtering
or any other technique that simply looks at the words available in the body
of the post. I am also curious to know if OL2003's junk filter even
includes looking at the text *within* the tags for HTML where spammers often
hide text (which are not valid HTML tags and why they then appear in the
view of the message). Microsoft has remained furtive on just how their spam
filtering works.
