spamprobe - Man Page

a bayesian spam filter


spamprobe [options] <command> [filename...]


SpamProbe can be used in conjunction with procmail or similar program to filter email.  SpamProbe uses a statistical algorithm to identify the key words and phrases in email and determine which emails are legitimate and which are spam.  The algorithm used by SpamProbe is based on an excellent article by Paul Graham.  He describes the basic idea and his results.  You can read his article here:

Command Line Usage

SpamProbe accepts a small set of commands and a growing set of options on the command line in addition to zero or more file names of mboxes. The general usage is:

 spamprobe [options] <command> [filename...]

The recognized options are:

-a char

   By default SpamProbe converts non-ascii characters (characters
   with the most significant bit set to 1) into the letter 'z'.  This
   is useful for lumping all Asian characters into a single word for
   easy recognition.  The -a option allows you to change the
   character to something else if you don't like the letter 'z' for
   some reason.


   Tells spamprobe to create the database directory if it does not
   already exist.  Normally spamprobe exits with a usage error if
   the database directory does not already exist.

-C number

   Tells SpamProbe to assign a default, somewhat neutral, probability
   to any term that does not have a weighted (good count doubled)
   count of at least number in the database.  This prevents terms
   which have been seen only a few times from having an unreasonable
   influence on the score of an email containing them.

   The default value is 5.  For example if number is 5 then in order
   for a term to use its calculated probability it must have been
   seen 3 times in good mails, or 2 times in good mails and once in
   spam, or 5 times in spam, or some other combination adding up to
   at least 5.

-d directory

   By default SpamProbe stores its database in a directory named
   .spamprobe under your home directory.  The -d option allows you to
   specify a different directory to use.  This is necessary if your
   home directory is NFS mounted for example.

   The directory name can be prefixed with a special code to force
   SpamProbe to use a particular type of data file format.  The type
   codes depend on how your copy of SpamProbe was compiled.  Defined
   types include:

     Example                   Description
     -d pbl:path               Forces the use of PBL data file.
     -d hash:path              Forces the use of an mmapped hash file.
     -d split:path             Forces the use of a hash file and ISAM
                               file (may provide better precision than
                               plain hash in some cases).

   The hash: option can also specify a desired file size in megabytes
   before the path.  For example -d hash:19:path would cause
   SpamProbe to use a 19 MB hash file.  The size must be in the range
   of 1-100.  The default hash file size is 16 MB.  Because hash
   files have a fixed size and capacity they should be cleaned
   relatively often using the cleanup command (see below) to prevent
   them from becoming full or being slowed by too many hash key

   Hash files provide better performance than either of the ISAM
   options (PBL or Berkeley DB).  However hash files do not store the
   original terms.  Only a 32 bit hash key is stored with each term.
   This prevents a user from exploring the terms in the database
   using the dump command to see what words are particularly spammy
   or hammy.

-D directory

   Tells SpamProbe to use the database in the specified directory
   (must be different than the one specified with the -d option) as a
   shared database from which to draw terms that are not defined in
   the user's own database.  This can be used to provide a baseline
   database shared by all users on a system (in the -D directory) and
   a private database unique to each user of the system
   ($HOME/.spamprobe or -d directory).

-g field_name

   Tells SpamProbe what header to look for previous score and message
   digest in.  Default is X-SpamProbe.  Field name is not case
   sensitive.  Used by all commands except receive.


   By default SpamProbe removes HTML markup from the text in emails
   to help avoid false positives.  The -h option allows you to
   override this behavior and force SpamProbe to include words from
   within HTML tags in its word counts.  Note that SpamProbe always
   counts any URLs in hrefs within tags whether -h is used or not.
   Use of this option is discouraged.  It can increase the rate of
   spam detection slightly but unless the user receives a significant
   amount of HTML emails it also tends to increase the number of
   false positives.

-H option

   By default SpamProbe only scans a meaningful subset of headers
   from the email message when searching for words to score.  The -H
   option allows the user to specify additional headers to scan.
   Legal values are "all", "nox", "none", or "normal".  "all" scans
   all headers, "nox" scans all headers except those starting with
   X-, "none" does not scan headers, and "normal" scans the normal
   set of headers.

   In addition to those values you can also explicitly add a header
   to the list of headers to process by adding the header name in
   lower case preceded by a plus sign.  Multiple headers can be
   specified by using multiple -H options.  For example, to include
   only the From and Received headers in your train command you could
   run spamprobe as follows:

     spamprobe -Hnone -H+from -H+received train

   You can also selectively ignore headers that would otherwise be
   processed by using -H-headername.  For example to process all
   headers except for Subject you could run spamprobe as follows:

     spamprobe -Hall -H-subject train

   To process the normal set of headers but also add the SpamAssassin
   header X-SpamStatus you could run spamprobe as follows:

     spamprobe -H+x-spam-status train

-l number

  Changes the spam probability threshold for emails from the default
  (0.7) to number.  The number must be a between 0 and 1.  Generally
  the value should be above 0.5 to avoid a high false positive rate.
  Lower numbers tend to produce more false positives while higher
  numbers tend to reduce accuracy.


   Forces SpamProbe to use mbox format for reading emails in receive
   mode.  Normally SpamProbe assumes that the input to receive mode
   contains a single message so it doesn't look for message breaks.


   Forces SpamProbe to treat the entire input as a single message.
   This ignores From lines and Content-Length headers in the input.

-o option_name

   Enables special options by name.  Currently the only special
   options are:

     -o graham

       Causes SpamProbe to emulate the filtering algorithm originally
       outlined in A Plan For Spam.

     -o honor-status-header

       Causes SpamProbe to ignore messages if they have a Status:
       header containing a capital D.  Some mail servers use this
       status to indicate a message that has been flagged for
       deletion but has not yet been purged from the file.

       DO NOT use this option with the receive or train command in
       your procmailrc file!  Doing so could allow spammers to bypass
       the filter.  This option is meant to be used with the
       train-spam and train-good commands in scripts that
       periodically update the database.

     -o honor-xstatus-header

       Causes SpamProbe to ignore messages if they have a X-Status:
       header containing a capital D.  Some mail servers use this
       status to indicate a message that has been flagged for
       deletion but has not yet been purged from the file.

       DO NOT use this option with the receive or train command in
       your procmailrc file!  Doing so could allow spammers to bypass
       the filter.  This option is meant to be used with the
       train-spam and train-good commands in scripts that
       periodically update the database.

     -o ignore-body

       Causes SpamProbe to ignore terms from the message body when
       computing a score.  This is not normally recommended but might
       be useful in conjunction with some other filter.  For example,
       the whitelist option (see below) implicitly ignores the
       message body.

     -o orig-score

       Causes SpamProbe to use its original scoring algorithm that
       produces excellent results but tends to generate scores of
       either 0 or 1 for all messages.

     -o suspicious-tags

       Causes SpamProbe to scan the contents of "suspicious" tags for
       tokens rather than simply throwing them out.  Currently only
       font tags are scanned but other tags may be added to this list
       in later versions.

     -o tokenized

       Causes SpamProbe to read tokens one per line rather than
       processing the input as mbox format.  This allows users to
       completely replace the standard spamprobe tokenizer if they
       wish and instead use some external program as a tokenizer.
       For example in your procmailrc file you could use:

        SCORE=| | /bin/spamprobe -o tokenized train

       In this mode SpamProbe considers a blank line to indicate the
       end of one message's tokens and the start of a new message's
       tokens.  SpamProbe computes a message digest based on the
       lines of text containing the tokens.

     -o whitelist

       Causes SpamProbe to use information from the email's headers
       to identify whether or not the email is from a legitimate
       correspondent.  The message body is ignored as are any never
       before seen terms and phrases in the headers.  This option can
       be used with the score command in a procmailrc file to use a
       bayesian white list in conjunction with some other filter or
       rule external to SpamProbe.

   The -o option can be used multiple times and all requested options
   will be applied.  Note that some options might conflict with each
   other in which case the last option would take precedence.

-p number

   Changes the maximum number of words per phrase.  Default value is
   two.  Increasing the limit improves accuracy somewhat but
   increases database size.  Experiments indicate that increasing
   beyond two is not worth the extra cost in space.

-P number

   Causes spamprobe to perform a purge of all terms with junk count
   less than or equal 2 after every number messages are processed.
   Using this option when classifying a large collection of spam can
   prevent the database from growing overly large at the cost of more
   processing time and possible loss of precision.

-r number

   Changes the number of times that a single word/phrase can occur
   in the top words array used to calculate the score for each
   message.  Allowing repeats reduces the number of words overall
   (since a single word occupies more than one slot) but allows words
   which occur frequently in the message to have a higher weight.
   Generally this is changed only for optimization purposes.


   Causes spamprobe to treat the input as a single message and to
   base its exit code on whether or not that message was spam.  The
   exit code will be 0 if the message was spam or 1 if the message
   was good.

-s number

   SpamProbe maintains an in memory cache of the words it has seen in
   previous messages to reduce disk I/O and improve performance.  By
   default the cache will contain the most recently accessed 2,500
   terms.  This number can be changed using the -s option.  Using a
   larger the cache size will cause SpamProbe to use more memory and,
   potentially, to perform less database I/O.

   A value of zero causes SpamProbe to use 100,000 as the limit which
   effectively means that the cache will only be flushed at program
   exit (unless you have really enormous mailbox files).  The cache
   doesn't affect receive, dump, or export but has a significant
   impact on the others.

   Causes SpamProbe to write out the top terms associated with each
   message in addition to its normal output.  Works with find-good,
   find-spam, and score.


   Tells SpamProbe to write debugging information to stderr.  This
   can be useful for debugging or for seeing which terms SpamProbe
   used to score each email.


   Prints version and copyright information and then exits.

-w number

   Changes the number of most significant words/phrases used by
   SpamProbe to calculate the score for each message.  Generally this
   is changed only for optimization purposes.


   Normally SpamProbe uses only a fixed number of top terms (as set
   by the -w command line option) when scoring emails.  The -x option
   can be used to allow the array to be extended past the max size if
   more terms are available with probabilities <= 0.1 or >= 0.9.


   An interesting variation on the scoring settings.  Equivalent to
   using "-w5 -r5 -x" so that generally only words with probabilites
   <= 0.1 or >= 0.9 are used and word frequencies in the email count
   heavily towards the score.  Tests have shown that this setting
   tends to be safer (fewer false positives) and have higher recall
   (proper classification of spams previously scored as spam)
   although its predictive power isn't quite as good as the default
   settings.  WARNING: This setting might work best with a fairly
   large corpus, it has not been tested with a small corpus so it
   might be very inaccurate with fewer than 1000 total messages.


   Assume traditional Berkeley mailbox format, ignoring any
   Content-Length: fields.


   Tells SpamProbe to ignore any characters with the most significant
   bit set to 1 instead of mapping them to the letter 'z'.


   Tells SpamProbe to store all characters even if their most
   significant bit is set to 1.

SpamProbe recognizes the following commands:

spamprobe help [command]

  With no arguments spamprobe lists all of the valid commands.
  If one or more commands are specified after the word help,
  spamprobe will print a more verbose description of each command.

spamprobe create-db

  If no database currently exists spamprobe will attempt to create
  one and then exit.  This can be used to bootstrap a new
  installation.  Strictly speaking this command is not necessary
  since the train-spam, train-good, and auto-train commands will also
  create a database if none already exists but some users like to
  create a database as a separate installation step.

spamprobe create-config

  Writes a new configuration file named spamprobe.hdl into the
  database directory (normally $HOME/.spamprobe).  Any existing
  configuration file will be overwritten so be sure to make a copy
  before invoking this command.

spamprobe receive [filename...]

  Tells SpamProbe to read its standard input (or a file specified
  after the receive command) and score it using the current
  databases.  Once the message has been scored the message is
  classified as either spam or non-spam and its word counts are
  written to the appropriate database.  The message's score is
  written to stdout along with a single word.  For example:

    SPAM 0.9999999 595f0150587edd7b395691964069d7af


    GOOD 0.0200000 595f0150587edd7b395691964069d7af

  The string of numbers and letters after the score is the message's
  "digest", a 32 character number which uniquely identifies the
  message.  The digest is used by SpamProbe to recognize messages
  that it has processed previously so that it can keep its word
  counts consistent if the message is reclassified.

  Using the -T option additionally lists the terms used to produce
  the score along with their counts (number of times they were found
  in the message).

spamprobe train [filename...]

  Functionally identical to receive except that the database is only
  modified if the message was "difficult" to classify.  In practice
  this can reduce the number of database updates to as little as 10%
  of messages received.

spamprobe score [filename...]

  Similar to receive except that the database is not modified in
  any way.

spamprobe summarize [filename...]

  Similar to score except that it prints a short summary and score
  for each message.  This can be useful when testing.  Using the -T
  option additionally lists the terms used to produce the score along
  with their counts (number of times they were found in the message).

spamprobe find-spam [filename...]

  Similar to score except that it prints a short summary and score
  for each message that is determined to be spam.  This can be useful
  when testing.  Using the -T option additionally lists the terms
  used to produce the score along with their counts (number of times
  they were found in the message).

spamprobe find-good [filename...]

  Similar to score except that it prints a short summary and score
  for each message that is determined to be good.  This can be useful
  when testing.  Using the -T option additionally lists the terms
  used to produce the score along with their counts (number of times
  they were found in the message).

spamprobe auto-train {SPAM|GOOD filename...}...

  Attempts to efficiently build a database from all of the named
  files.  You may specify one or more file of each type.  Prior to
  each set of file names you must include the word SPAM or GOOD to
  indicate what type of mail is contained in the files which follow
  on the command line.

  The case of the SPAM and GOOD keywords is important.  Any number of
  file names can be specified between the keywords.  The command line
  format is very flexible.  You can even use a find command in
  backticks to process whole directory trees of files. For example:

    spamprobe auto-train SPAM spams/* GOOD `find hams -type f`

  SpamProbe pre-scans the files to determine how many emails of each
  type exist and then trains on hams and spams in a random sequence
  that balances the inflow of each type so that the train command can
  work most effectively.  For example if you had 400 hams and 400
  spams, auto-train will generally process one spam, then one ham,
  etc.  If you had 4000 spams and 400 hams then auto-train will
  generally process 10 spams, then one ham, etc.

  Since this command will likely take a long time to run it is often
  desireable to use it with the -v option to see progress information
  as the messages are processed.

    spamprobe -v auto-train SPAM spams/* GOOD hams/*

spamprobe good [filename...]

  Scans each file (or stdin if no file is specified) and reclassifies
  every email in the file as non-spam.  The databases are updated
  appropriately.  Messages previously classified as good (recognized
  using their MD5 digest or message ids) are ignored.  Messages
  previously classified as spam are reclassified as good.

spamprobe train-good [filename...]

  Functionally identical to "good" command except that it only
  updates the database for messages that are either incorrectly
  classified (i.e. classified as spam) or are "difficult" to
  classify.  In practice this can reduce amount of database updates
  to as little as 10% of messages.

spamprobe spam [filename...]

  Scans each file (or stdin if no file is specified) and reclassifies
  every email in the file as spam.  The databases are updated
  appropriately.  Messages previously classified as spam (recognized
  using their MD5 digest of message ids) are ignored.  Messages
  previously classified as good are reclassified as spam.

spamprobe train-spam [filename...]

  Functionally identical to "spam" command except that it only
  updates the database for messages that are either incorrectly
  classified (i.e. classified as good) or are "difficult" to
  classify.  In practice this can reduce amount of database updates
  to as little as 10% of messages.

spamprobe remove [filename...]

  Scans each file (or stdin if no file is specified) and removes its
  term counts from the database.  Messages which are not in the
  database (recognized using their MD5 digest of message ids) are

spamprobe cleanup [ junk_count [ max_age ] ]...

  Scans the database and removes all terms with junk_count or less
  (default 2) which have not had their counts modified in at least
  max_age days (default 7).  You can specify multiple count/age pairs
  on a single command line but must specify both a count and an age
  for all but the last count.  This should be run periodically to
  keep the database from growing endlessly.

  For my own email I use cron to run the cleanup command every day
  and delete all terms with count of 2 or less that have not been
  modified in the last two weeks.  Here is the excerpt from my

      3 0 * * * /home/brian/bin/spamprobe cleanup 2 14

  Alternatively you might want to use a much higher count (1000 in
  this example) for terms that have not been seen in roughly six

      3 0 * * * /home/brian/bin/spamprobe cleanup 1000 180 2 14

  Because of the way that PBL and BerkeleyDB work the database file
  will not actually shrink, but newly added terms will be able to use
  the space previously occupied by any removed terms so that the
  file's growth should be significantly slower if this command is

  To actually shrink the database you can build a new one using the
  BerkeleyDB utility programs db_dump and db_load (Berkeley DB only)
  or the spamprobe import and export commands (either database
  library).  For example:

      cd ~
      mkdir new.spamprobe
      spamprobe export | spamprobe -d new.spamprobe import
      mv .spamprobe old.spamprobe
      mv new.spamprobe .spamprobe

  The -P option can also be used to limit the rate of growth of the
  database when importing a large number of emails.  For example if
  you want to classify 1000 emails and want SP to purge rare terms
  every 100 messages use a command such as:

    spamprobe -P 100 good goodmailboxname

  Using -P slows down the classification but can avoid the need to
  use the db_dump trick.  Using -P only makes sense when classifying
  a large number of messages.

spamprobe purge [ junk_count ]

  Similar to cleanup but forces the immediate deletion of all terms
  with total count less than junk_count (default is 2) no matter how
  long it has been since they were modified (i.e. even if they were
  just added today). This could be handy immediately after
  classifying a large mailbox of historical spam or good email to
  make room for the next batch.

spamprobe purge-terms regex

  Similar to purge except that it removes from the database all terms
  which match the specified regular expression.  Be careful with this
  command because it could remove many more terms than you expect.
  Use dump with the same regex before running this command to see
  exactly what will be deleted.

spamprobe edit-term term good_count spam_count

  Can be used to specifically set the good and spam counts of a term.
  Whether this is truly useful is doubtful but it is provided for
  completeness sake.  For example it could be used to force a
  particular word to be very spammy or very good:

      spamprobe edit-term nigeria 0 1000000
      spamprobe edit-term burton  10000000 0

spamprobe dump [ regex ]

  Prints the contents of the word counts database one word per line
  in human readable format with spam probability, good count, spam
  count, flags, and word in columns separated by whitespace.  PBL and
  Berkeley DB sort terms alphabetically.  The standard unix sort
  command can be used to sort the terms as desired.  For example to
  list all words from "most good" to "least good" use this command:

      spamprobe dump | sort -k 1n -k 2nr

  To list all words from "most spammy" to "least spammy" use this

      spamprobe dump | sort -k 1nr -k 3nr

  Optionally you can specify a regular expression.  If specified
  SpamProbe will only dump terms matching the regular expression.
  For example:

      spamprobe dump 'finance'
      spamprobe dump 'bfinanceb'
      spamprobe dump 'HSubject_.*finance'

spamprobe tokenize [ filename ]

  Prints the tokens found in the file one word per line in human
  readable format with spam probability, good count, spam count,
  message count, and word in columns separated by whitespace.  Terms
  are listed in the order in which they were encountered in the
  message.  The standard unix sort command can be used to sort the
  terms as desired.  For example to list all words from "most good"
  to "least good" use this command:

      spamprobe tokenize filename | sort -k 1n -k 2nr

  To list all words from "most spammy" to "least spammy" use this

      spamprobe tokenize filename | sort -k 1nr -k 3nr

spamprobe export

  Similar to the dump command but prints the counts and words in a
  comma separated format with the words surrounded by double quotes.
  This can be more useful for importing into some databases.

spamprobe import

  Reads the specified files which must contain export data written by
  the export command.  The terms and counts from this file are added
  to the database.  This can be used to convert a database from a
  prior version.

spamprobe exec command

  Obtains an exclusive lock on the database and then executes the
  command using system(3).  If multiple arguments are given after
  "exec" they are combined to form the command to be executed.  This
  command can be used when you want to perform some operation on the
  database without interference from incoming mail.  For example, to
  back up your .spamprobe directory using tar you could do something
  like this:

      spamprobe exec tar cf spamprobe-data.tar.gz .spamprobe

  If you simply want to hold the lock while interactively running
  commands in a different xterm you could use "spamprobe exec read".
  The linux read program simply reads a line of text from your
  terminal so the lock would effectively be held until you pressed
  the enter key.  Another option would be to use a shell as the
  command and type the commands into that shell:

      spamprobe /bin/bash

  Be careful not to run spamprobe in the shell though since the
  spamprobe in the shell will wind up deadlocked waiting for the
  spamprobe running the exec command to release its lock.

spamprobe exec-shared command

  Same as exec except that a shared lock is used.  This may be more
  appropriate if you are backing up your database since operations
  like score (but not train or receive) could still be performed on
  the database while the backup was running.

Setup of Spamprobe for Users

Once you have a spamprobe executable copy it to someplace in your PATH so that procmail can find it.  Then create a directory for SpamProbe to store its databases in.  By default SpamProbe wants to use the directory ~/.spamprobe.  You must create this directory manually in order to run SpamProbe or else specify some other directory using the -d option.  Something like this should suffice:

 mkdir ~/.spamprobe

SpamProbe can use either the PBL or Berkeley DB library for its databases.  Both are fast on local file systems but very slow over NFS.  Please ensure that your spamprobe directory is on a local file system to ensure good performance.

Notes Using Hash Database

SpamProbe can use a simple, fixed size hash data file as an alternative to PBL or BDB.  There are two advantages to the hash format.  The first is speed.  In my experiments the hash file format is around 2x the speed of PBL (ranged from 1.8x to 3.5x). The second advantage is that the hash data file size is fixed.  You choose a size when you create the file and it never changes.  File size can be anywhere from 1-100 MB. You need to choose a size large enough to hold your terms with room to spare.  More on that later.

The hash file format also has significant disadvantages.  Becuase the file size is fixed you must monitor the file to ensure that it does not become overly full.  When the file becomes more than half full performance will suffer.  Also the hash format does not store original terms so you cannot use the dump command to learn what terms are spammy or hammy in your database.  Finally, the hash format is imprecise.  Hash collisions can cause the counts from different terms to be mixed together which can reduce accuracy.

To create a hash data file you add a prefix to the directory name in the -d command line option.  You can specify just the directory like this:

 spamprobe -d hash:$HOME/.spamprobe

or you can add a size in megabytes for the file like this:

 spamprobe -d hash:42:$HOME/.spamprobe

The size is only used when a file is first created.  SP auto detects the size of an existing hash file.  You need to allow enough space for twice as many terms as you are likely to have in your file.  In my database I have 2.2 million terms.  That required a database of are 53 MB.  SP uses 12 bytes per term in the hash file so you can estimate the file size you'll need by multiplying the number of terms by 24.

The hash format does not store the original terms.  Instead it stores the 32 bit hash code for each term.  You can do just about anything with a hash file that you could with a PBL file including import/export, edit-term, cleanup, purge, etc.  You can use export your PBL database and import it to build a hash file (note that you cannot go the other direction) and you can export one hash file and import into a new one to enlarge your file.

Maildir Format

SpamProbe will accept a maildir directory name anywhere that an Mbox or MBX file name can be specified.  When SpamProbe encounters a Maildir mailbox (directory) name it will automatically process all of the non-hidden files in the cur and new subdirectories of the mailbox. There is no need to individually specify these subdirectories.

Getting Started

SpamProbe is not a stand alone mail filter.  It doesn't sort your mail or split it into different mailboxes.  Instead it relies on some other program such as procmail to actually file your mail for you.  What SpamProbe does do is track the word counts in good and spam emails and generate a score for each email that indicates whether or not it is likely to be spam.  Scores range from 0 to 1 with any score of 0.9 or higher indicating a probable spam.

Personally I use SpamProbe with procmail to filter my incoming email into mail boxes.  I have procmail score each inbound email using SpamProbe and insert a special header into each email containing its score.  Then I have procmail move spams into a special mailbox.

No spam filter is perfect and SpamProbe sometimes makes mistakes.  To correct those mistakes I have a special mailbox that I put undetected spams into.  I run SpamProbe periodically and have it reclassify any emails in that mailbox as spam so that it will make a better guess the next time around.

This is not a procmail primer.  You will need to ensure that you have procmail and formail installed before you can use this technique. Also I recommend that you read the procmail documentation so that you can fully understand this example and adapt it to your own needs. That having been said, my .procmailrc file looks like this:


   :0 c

   SCORE=| /home/brian/bin/spamprobe train
   :0 wf
   | formail -I "X-SpamProbe: $SCORE"
   :0 a:
   *^X-SpamProbe: SPAM

I use IMAP to fetch my email so my mailboxes all live in a directory named IMAP on my mail server.

NOTE: The first stanza copies all incoming emails into a special mbox called saved.  SpamProbe IS BETA SOFTWARE and though it works well for me it is possible that it could somehow lose emails.  Caution is always a good idea.  That having been said, with the procmailrc file as shown above the worst that could happen if SpamProbe crashes is that the email would not be scored properly and procmail would deliver it to your inbox.  Of course if procmail crashes all bets are off.

The second stanza runs spamprobe in "train" mode to score the email, classify it as either spam or good, and possibly update the database. The train command tries to minimize the number of database updates by only updating the database with terms from an incoming message if there was insufficient confidence in the message's score.  The train command always updates the database on the first 1500 of each type received.  This ensures that sufficient email is classified to allow the filter to operate reliably.

The next stanza runs formail to add a custom header to the email containing the SpamProbe score.  The final stanza uses the contents of the custom header to file detected spams into a special mbox named spamprobe.

As an alternative to using the train command, you can run spamprobe in "receive" mode.  In that mode SpamProbe scores the email and then classifies it as either spam or good based on the score.  It always automatically adds the word counts for the email to the appropriate database.  This is essentially like running in score mode followed immediately by either spam or good mode.  It produces more database I/O and a bigger database but ensures that every message has its terms reflected in the database.  Personally I use train mode.  A sample procmailrc file using the receive command looks like this:


   :0 c

   SCORE=| /home/brian/bin/spamprobe receive
   :0 wf
   | formail -I "X-SpamProbe: $SCORE"
   :0 a:
   *^X-SpamProbe: SPAM

Making Corrections

SpamProbe is not perfect.  It is able to detect over 99% of the spams that I receive but some still slip through.  To correct these missed emails I run SpamProbe periodically and have it scan a special mbox. Since I use IMAP to retrieve my emails I can simply drop undetected spams into this mbox from my mail client.  If you use POP or some other system then you will need to find a way get the undetected spams into a mbox that spamprobe can see.

Periodically I run a script that scans three special mboxes to correct errors in judgment:



   spamprobe remove $IMAPDIR/remove
   spamprobe good $IMAPDIR/nonspam
   spamprobe spam $IMAPDIR/spam
   spamprobe train-spam $IMAPDIR/spamprobe

From this example you can see that I use three special mboxes to make corrections.  I copy emails that I don't want spamprobe to store into the remove mbox.  This is useful if you receive email from a friend or colleague that looks like spam and you don't want it to dilute the effectiveness of the terms it contains.

Undetected spams go into the spam mbox.  SpamProbe will reclassify those emails as spam and correct its database accordingly.  Note that doing this does not guarantee that the spam will always be scored as spam in the future.  Some spams are too bland to detect perfectly. Fortunately those are very rare.

The nonspam mbox is for any false positives.  These are always possible and it is important to have a way to reclassify them when they do occur.

If you are using receive mode rather than train mode then the above script can be modified to remove the train-spam line. For example:



   spamprobe remove $IMAPDIR/remove
   spamprobe good $IMAPDIR/nonspam
   spamprobe spam $IMAPDIR/spam

Finally you'll need to build a starting database.  Since SpamProbe relies on word counts from past emails it requires a decent sized database to be accurate.  To build the database select some of your mboxes containing past emails.  Ideally you should have one mbox of spams and one or more of non-spams.  If you don't have any spams handy then don't worry, SpamProbe will gradually become more accurate as you receive more spams.  Expect a fairly high false negative (i.e. missed spams) rate as you first start using SpamProbe.

To import your starting messages use commands such as these.  The example assumes that you have non-spams stored in a file named mbox in your home directory and some spams stored in a file named nasty-spams. Replace these names with real ones.

 spamprobe good ~/mbox
 spamprobe spam ~/nasty-spams

See Also



December 2005 Version 1.4 SpamProbe