>...
>Date: Wed, 16 Mar 2005 09:38:13 -0000 (GMT)
>Subject: Re: Is there such a test?
>From: "Mike Spamassassin" <[EMAIL PROTECTED]>
>
>I'd take that bet.
>While you are almost certainly correct with the likes of those who
>subscribe to this group, who often have multiple email addresses,
>out there in [EMAIL PROTECTED] land, and hotmail world, most people have a
>single
>email address strongly related to their name.
>
>Back to the original question:
>Regardless of whether anyone thinks it is a good test or not, has anyone
>yet created such a test?
>
>> Mike Spamassassin wrote:
>>
>>>Point taken, but I still think it would be a valid test.
>>>Like all SpamAssassin tests it should only be one of many indicators.
>>>
>>
>> No, not really. There's a minimum useful S/O ratio for spam rules.
>>
>> I'd bet $5.00 that this rule would have a S/O under 0.80 in the
>> corpus.(ie: no more 80% of it's hits were spam, and at least 20% were ham)
>>
>>
>
>
>
Mike,
I think you are probably correct for exactly the group you describe;
Individual users (i.e. "in [EMAIL PROTECTED] land, and hotmail world"). But,
you'd
get killed with corporate accounts. I haven't made it to London is quite
a while, but if you change the bet to a pint, I'll take it too. (BTW. MS
passports, unlike normal Hotmail/MSN accounts would fail very often also,
and "techies" often have "cute" account names, so anyone who does a lot
of technical discussions would might have have problems).
I think the only mail I normally see that wouldn't fail your test
is a few mailing list posters (and many counter examples have already been
listed for you, I could give you another couple of dozen I get nearly every
day) and relatives (and many of them are weird; e.g. "The LastName Household"
with accounts like [EMAIL PROTECTED]/com are common - It used to be
on one of their example pages for "family" accounts with multiple mailboxes).
And it would be hard to pick out the single letter of the account name that
matches the first letter of the "Lastname" - in fact, to me, initials followed
by digits "seem" to be a spammer unless I know the person (but I doubt that
does any better than 50%, or even that well) - now names followed, by a year
might meet the 80% test (ex. [EMAIL PROTECTED] or something like an account
[EMAIL PROTECTED]). Didn't we have one of those, asking people
to accept mail from invalid domains, on the list a couple of weeks ago?
Paul Shupak
[EMAIL PROTECTED]
