Re: [Spamprobe-users] spamprobe remove

[Brian B. <bb...@us...>]

--On Saturday, January 31, 2004 12:51 AM -0600 Graham Toal 
<gt...@gt...> wrote:
>> --On Wednesday, January 28, 2004 7:09 PM -0600 root <ro...@gt...>
>> wrote:
>> > If I have a single shared database (which I do) and many users (which I
>> > will), and a few of those users decide to mark some item (such as an
>> > email from the president of the university) as spam, will any of the
>> > following be true?:
>> The shared database is read only.  Each user also has a local database
>> that  overrides matching records in the shared database.  For example
>> imagine  this scenario:
>
> I'm sorry, I wasn't clear: I'm not referring to spamprobe's concept of
> a shared database; I mean that mail for all users is being processed
> through a single choke-point by one username and a single database.  I.e.
> all training is occuring on one single database.  There are no personal
> databases for the individual users.  This shared database (perhaps it
> would be less confusing if I start calling it a 'communal database') is
> not read-only.  It's updated by everyone. (Or it will be.  As yet there is
> no user feedback, it works entirely on self-training and spamtraps so far)
> Anyway, what one person feeds back *will* affect everyone else.

Oh. Sorry.  Well at least my other post can serve as a good explanation for 
others. :-)

In your case your accuracy can suffer if people have different opinions 
about the same email.  You are correct that the last one to reclassify a 
particular email will win.  SP is specifically designed that way so that 
people can correct their own database's results quickly.

SP includes the From: and To: in its calculation of the digest.  If the 
email in question had the recipient's email address in the To: or From: 
then it would not override a classification by another person.  However if 
they went to a mailing list (so all had the same From and To) then they 
will clash and the last person to reclassify would "win". You could always 
modify SP to have it include the Received headers in the digest.  To do so 
change getMD5Digest() in MimeMessageReader.cc to the following:

const string &MimeMessageReader::getMD5Digest()
{
  if (m_md5.isRunning()) {
    addFieldToDigest("date");
    addFieldToDigest("from");
    addFieldToDigest("to");
    addFieldToDigest("subject");
    addFieldToDigest("received");
    m_md5.stop();
  }
  return m_md5.asString();
}



All the best,
++Brian