1 ;;; spam-stat.el --- detecting spam based on statistics
3 ;; Copyright (C) 2002-2015 Free Software Foundation, Inc.
5 ;; Author: Alex Schroeder <alex@gnu.org>
7 ;; URL: http://www.emacswiki.org/cgi-bin/wiki.pl?SpamStat
9 ;; This file is part of GNU Emacs.
11 ;; GNU Emacs is free software: you can redistribute it and/or modify
12 ;; it under the terms of the GNU General Public License as published by
13 ;; the Free Software Foundation, either version 3 of the License, or
14 ;; (at your option) any later version.
16 ;; GNU Emacs is distributed in the hope that it will be useful,
17 ;; but WITHOUT ANY WARRANTY; without even the implied warranty of
18 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 ;; GNU General Public License for more details.
21 ;; You should have received a copy of the GNU General Public License
22 ;; along with GNU Emacs. If not, see <http://www.gnu.org/licenses/>.
26 ;; This implements spam analysis according to Paul Graham in "A Plan
27 ;; for Spam". The basis for all this is a statistical distribution of
28 ;; words for your spam and non-spam mails. We need this information
29 ;; in a hash-table so that the analysis can use the information when
30 ;; looking at your mails. Therefore, before you begin, you need tons
31 ;; of mails (Graham uses 4000 non-spam and 4000 spam mails for his
34 ;; The main interface to using spam-stat, are the following functions:
36 ;; `spam-stat-buffer-is-spam' -- called in a buffer, that buffer is
37 ;; considered to be a new spam mail; use this for new mail that has
38 ;; not been processed before
40 ;; `spam-stat-buffer-is-non-spam' -- called in a buffer, that buffer
41 ;; is considered to be a new non-spam mail; use this for new mail that
42 ;; has not been processed before
44 ;; `spam-stat-buffer-change-to-spam' -- called in a buffer, that
45 ;; buffer is no longer considered to be normal mail but spam; use this
46 ;; to change the status of a mail that has already been processed as
49 ;; `spam-stat-buffer-change-to-non-spam' -- called in a buffer, that
50 ;; buffer is no longer considered to be spam but normal mail; use this
51 ;; to change the status of a mail that has already been processed as
54 ;; `spam-stat-save' -- save the hash table to the file; the filename
55 ;; used is stored in the variable `spam-stat-file'
57 ;; `spam-stat-load' -- load the hash table from a file; the filename
58 ;; used is stored in the variable `spam-stat-file'
60 ;; `spam-stat-score-word' -- return the spam score for a word
62 ;; `spam-stat-score-buffer' -- return the spam score for a buffer
64 ;; `spam-stat-split-fancy' -- for fancy mail splitting; add
65 ;; the rule (: spam-stat-split-fancy) to `nnmail-split-fancy'
67 ;; This requires the following in your ~/.gnus file:
69 ;; (require 'spam-stat)
74 ;; Typical test will involve calls to the following functions:
76 ;; Reset: (spam-stat-reset)
77 ;; Learn spam: (spam-stat-process-spam-directory "~/Mail/mail/spam")
78 ;; Learn non-spam: (spam-stat-process-non-spam-directory "~/Mail/mail/misc")
79 ;; Save table: (spam-stat-save)
80 ;; File size: (nth 7 (file-attributes spam-stat-file))
81 ;; Number of words: (hash-table-count spam-stat)
82 ;; Test spam: (spam-stat-test-directory "~/Mail/mail/spam")
83 ;; Test non-spam: (spam-stat-test-directory "~/Mail/mail/misc")
84 ;; Reduce table size: (spam-stat-reduce-size)
85 ;; Save table: (spam-stat-save)
86 ;; File size: (nth 7 (file-attributes spam-stat-file))
87 ;; Number of words: (hash-table-count spam-stat)
88 ;; Test spam: (spam-stat-test-directory "~/Mail/mail/spam")
89 ;; Test non-spam: (spam-stat-test-directory "~/Mail/mail/misc")
91 ;;; Dictionary Creation:
93 ;; Typically, you will filter away mailing lists etc. using specific
94 ;; rules in `nnmail-split-fancy'. Somewhere among these rules, you
95 ;; will filter spam. Here is how you would create your dictionary:
97 ;; Reset: (spam-stat-reset)
98 ;; Learn spam: (spam-stat-process-spam-directory "~/Mail/mail/spam")
99 ;; Learn non-spam: (spam-stat-process-non-spam-directory "~/Mail/mail/misc")
100 ;; Repeat for any other non-spam group you need...
101 ;; Reduce table size: (spam-stat-reduce-size)
102 ;; Save table: (spam-stat-save)
106 ;; Speed it up. Integrate with Gnus such that it uses spam and expiry
107 ;; marks to call the appropriate functions when leaving the summary
108 ;; buffer and saves the hash table when leaving Gnus. More testing:
109 ;; More mails, disabling SpamAssassin, double checking algorithm, find
110 ;; improved algorithm.
114 ;; Ted Zlatanov <tzz@lifelogs.com>
115 ;; Jesper Harder <harder@myrealbox.com>
116 ;; Dan Schmidt <dfan@dfan.org>
117 ;; Lasse Rasinen <lrasinen@iki.fi>
118 ;; Milan Zamazal <pdm@zamazal.org>
123 (require 'mail-parse
)
125 (defvar gnus-original-article-buffer
)
127 (defgroup spam-stat nil
128 "Statistical spam detection for Emacs.
129 Use the functions to build a dictionary of words and their statistical
130 distribution in spam and non-spam mails. Then use a function to determine
131 whether a buffer contains spam or not."
135 (defcustom spam-stat-file
"~/.spam-stat.el"
136 "File used to save and load the dictionary.
137 See `spam-stat-to-hash-table' for the format of the file."
141 (defcustom spam-stat-unknown-word-score
0.2
142 "The score to use for unknown words.
143 Also used for words that don't appear often enough."
147 (defcustom spam-stat-max-word-length
15
148 "Only words shorter than this will be considered."
152 (defcustom spam-stat-max-buffer-length
10240
153 "Only the beginning of buffers will be analyzed.
154 This variable says how many characters this will be."
158 (defcustom spam-stat-split-fancy-spam-group
"mail.spam"
159 "Name of the group where spam should be stored.
160 If `spam-stat-split-fancy' is used in fancy splitting rules. Has
161 no effect when spam-stat is invoked through spam.el."
165 (defcustom spam-stat-split-fancy-spam-threshold
0.9
166 "Spam score threshold in spam-stat-split-fancy."
170 (defcustom spam-stat-washing-hook nil
171 "Hook applied to each message before analysis."
175 (defcustom spam-stat-score-buffer-user-functions nil
176 "List of additional scoring functions.
177 Called one by one on the buffer.
179 If all of these functions return non-nil answers, these numerical
180 answers are added to the computed spam stat score on the buffer. If
181 you defun such functions, make sure they don't return the buffer in a
182 narrowed state or such: use, for example, `save-excursion'. Each of
183 your functions is also passed the initial spam-stat score which might
186 Also be careful when defining such functions. If they take a long
187 time, they will slow down your mail splitting. Thus, if the buffer is
188 large, don't forget to use smaller regions, by wrapping your work in,
189 say, `with-spam-stat-max-buffer-size'."
193 (defcustom spam-stat-process-directory-age
90
194 "Max. age of files to be processed in directory, in days.
195 When using `spam-stat-process-spam-directory' or
196 `spam-stat-process-non-spam-directory', only files that have
197 been touched in this many days will be considered. Without
198 this filter, re-training spam-stat with several thousand messages
199 will start to take a very long time."
203 (defvar spam-stat-last-saved-at nil
204 "Time stamp of last change of spam-stat-file on this run")
206 (defvar spam-stat-syntax-table
207 (let ((table (copy-syntax-table text-mode-syntax-table
)))
208 (modify-syntax-entry ?-
"w" table
)
209 (modify-syntax-entry ?_
"w" table
)
210 (modify-syntax-entry ?.
"w" table
)
211 (modify-syntax-entry ?
! "w" table
)
212 (modify-syntax-entry ??
"w" table
)
213 (modify-syntax-entry ?
+ "w" table
)
215 "Syntax table used when processing mails for statistical analysis.
216 The important part is which characters are word constituents.")
218 (defvar spam-stat-dirty nil
219 "Whether the spam-stat database needs saving.")
221 (defvar spam-stat-buffer nil
222 "Buffer to use for scoring while splitting.
223 This is set by hooking into Gnus.")
225 (defvar spam-stat-buffer-name
" *spam stat buffer*"
226 "Name of the `spam-stat-buffer'.")
228 (defvar spam-stat-coding-system
229 (if (mm-coding-system-p 'emacs-mule
) 'emacs-mule
'raw-text
)
230 "Coding system used for `spam-stat-file'.")
234 (defun spam-stat-store-current-buffer ()
235 "Store a copy of the current buffer in `spam-stat-buffer'."
236 (let ((buf (current-buffer)))
237 (with-current-buffer (get-buffer-create spam-stat-buffer-name
)
239 (insert-buffer-substring buf
)
240 (setq spam-stat-buffer
(current-buffer)))))
242 (defun spam-stat-store-gnus-article-buffer ()
243 "Store a copy of the current article in `spam-stat-buffer'.
244 This uses `gnus-article-buffer'."
245 (with-current-buffer gnus-original-article-buffer
246 (spam-stat-store-current-buffer)))
248 ;; Data -- not using defstruct in order to save space and time
250 (defvar spam-stat
(make-hash-table :test
'equal
)
251 "Hash table used to store the statistics.
252 Use `spam-stat-load' to load the file.
253 Every word is used as a key in this table. The value is a vector.
254 Use `spam-stat-ngood', `spam-stat-nbad', `spam-stat-good',
255 `spam-stat-bad', and `spam-stat-score' to access this vector.")
257 (defvar spam-stat-ngood
0
258 "The number of good mails in the dictionary.")
260 (defvar spam-stat-nbad
0
261 "The number of bad mails in the dictionary.")
263 (defvar spam-stat-error-holder nil
264 "A holder for condition-case errors while scoring buffers.")
266 (defsubst spam-stat-good
(entry)
267 "Return the number of times this word belongs to good mails."
270 (defsubst spam-stat-bad
(entry)
271 "Return the number of times this word belongs to bad mails."
274 (defsubst spam-stat-score
(entry)
275 "Set the score of this word."
278 spam-stat-unknown-word-score
))
280 (defsubst spam-stat-set-good
(entry value
)
281 "Set the number of times this word belongs to good mails."
282 (aset entry
0 value
))
284 (defsubst spam-stat-set-bad
(entry value
)
285 "Set the number of times this word belongs to bad mails."
286 (aset entry
1 value
))
288 (defsubst spam-stat-set-score
(entry value
)
289 "Set the score of this word."
290 (aset entry
2 value
))
292 (defsubst spam-stat-make-entry
(good bad
)
293 "Return a vector with the given properties."
294 (let ((entry (vector good bad nil
)))
295 (spam-stat-set-score entry
(spam-stat-compute-score entry
))
300 (defun spam-stat-compute-score (entry)
301 "Compute the score of this word. 1.0 means spam."
302 ;; promote all numbers to floats for the divisions
303 (let* ((g (* 2.0 (spam-stat-good entry
)))
304 (b (float (spam-stat-bad entry
))))
307 ((= 0 spam-stat-ngood
)
309 ((= 0 spam-stat-nbad
)
313 (min .99 (/ (/ b spam-stat-nbad
)
314 (+ (/ g spam-stat-ngood
)
315 (/ b spam-stat-nbad
)))))))))
319 (defmacro with-spam-stat-max-buffer-size
(&rest body
)
320 "Narrow the buffer down to the first 4k characters, then evaluate BODY."
322 (when (> (- (point-max)
324 spam-stat-max-buffer-length
)
325 (narrow-to-region (point-min)
326 (+ (point-min) spam-stat-max-buffer-length
)))
329 (defun spam-stat-buffer-words ()
330 "Return a hash table of words and number of occurrences in the buffer."
331 (run-hooks 'spam-stat-washing-hook
)
332 (with-spam-stat-max-buffer-size
333 (with-syntax-table spam-stat-syntax-table
334 (goto-char (point-min))
335 (let ((result (make-hash-table :test
'equal
))
337 (while (re-search-forward "\\w+" nil t
)
338 (setq word
(match-string-no-properties 0)
339 count
(1+ (gethash word result
0)))
340 (when (< (length word
) spam-stat-max-word-length
)
341 (puthash word count result
)))
344 (defun spam-stat-buffer-is-spam ()
345 "Consider current buffer to be a new spam mail."
346 (setq spam-stat-nbad
(1+ spam-stat-nbad
))
349 (let ((entry (gethash word spam-stat
)))
351 (spam-stat-set-bad entry
(+ count
(spam-stat-bad entry
)))
352 (setq entry
(spam-stat-make-entry 0 count
)))
353 (spam-stat-set-score entry
(spam-stat-compute-score entry
))
354 (puthash word entry spam-stat
)))
355 (spam-stat-buffer-words))
356 (setq spam-stat-dirty t
))
358 (defun spam-stat-buffer-is-non-spam ()
359 "Consider current buffer to be a new non-spam mail."
360 (setq spam-stat-ngood
(1+ spam-stat-ngood
))
363 (let ((entry (gethash word spam-stat
)))
365 (spam-stat-set-good entry
(+ count
(spam-stat-good entry
)))
366 (setq entry
(spam-stat-make-entry count
0)))
367 (spam-stat-set-score entry
(spam-stat-compute-score entry
))
368 (puthash word entry spam-stat
)))
369 (spam-stat-buffer-words))
370 (setq spam-stat-dirty t
))
372 (autoload 'gnus-message
"gnus-util")
374 (defun spam-stat-buffer-change-to-spam ()
375 "Consider current buffer no longer normal mail but spam."
376 (setq spam-stat-nbad
(1+ spam-stat-nbad
)
377 spam-stat-ngood
(1- spam-stat-ngood
))
380 (let ((entry (gethash word spam-stat
)))
382 (gnus-message 8 "This buffer has unknown words in it")
383 (spam-stat-set-good entry
(- (spam-stat-good entry
) count
))
384 (spam-stat-set-bad entry
(+ (spam-stat-bad entry
) count
))
385 (spam-stat-set-score entry
(spam-stat-compute-score entry
))
386 (puthash word entry spam-stat
))))
387 (spam-stat-buffer-words))
388 (setq spam-stat-dirty t
))
390 (defun spam-stat-buffer-change-to-non-spam ()
391 "Consider current buffer no longer spam but normal mail."
392 (setq spam-stat-nbad
(1- spam-stat-nbad
)
393 spam-stat-ngood
(1+ spam-stat-ngood
))
396 (let ((entry (gethash word spam-stat
)))
398 (gnus-message 8 "This buffer has unknown words in it")
399 (spam-stat-set-good entry
(+ (spam-stat-good entry
) count
))
400 (spam-stat-set-bad entry
(- (spam-stat-bad entry
) count
))
401 (spam-stat-set-score entry
(spam-stat-compute-score entry
))
402 (puthash word entry spam-stat
))))
403 (spam-stat-buffer-words))
404 (setq spam-stat-dirty t
))
406 ;; Saving and Loading
408 (defun spam-stat-save (&optional force
)
409 "Save the `spam-stat' hash table as lisp file.
410 With a prefix argument save unconditionally."
412 (when (or force spam-stat-dirty
)
413 (let ((coding-system-for-write spam-stat-coding-system
))
414 (with-temp-file spam-stat-file
415 (let ((standard-output (current-buffer)))
416 (insert (format ";-*- coding: %s; -*-\n" spam-stat-coding-system
))
417 (insert (format "(setq spam-stat-ngood %d spam-stat-nbad %d
418 spam-stat (spam-stat-to-hash-table '(" spam-stat-ngood spam-stat-nbad
))
419 (maphash (lambda (word entry
)
421 (spam-stat-good entry
)
422 (spam-stat-bad entry
))))
425 (message "Saved %s." spam-stat-file
)
426 (setq spam-stat-dirty nil
427 spam-stat-last-saved-at
(nth 5 (file-attributes spam-stat-file
)))))
429 (defun spam-stat-load ()
430 "Read the `spam-stat' hash table from disk."
431 ;; TODO: maybe we should warn the user if spam-stat-dirty is t?
432 (let ((coding-system-for-read spam-stat-coding-system
))
433 (cond (spam-stat-dirty (message "Spam stat not loaded: spam-stat-dirty t"))
434 ((or (not (boundp 'spam-stat-last-saved-at
))
435 (null spam-stat-last-saved-at
)
436 (not (equal spam-stat-last-saved-at
437 (nth 5 (file-attributes spam-stat-file
)))))
439 (load-file spam-stat-file
)
440 (setq spam-stat-dirty nil
441 spam-stat-last-saved-at
442 (nth 5 (file-attributes spam-stat-file
)))))
443 (t (message "Spam stat file not loaded: no change in disk.")))))
445 (defun spam-stat-to-hash-table (entries)
446 "Turn list ENTRIES into a hash table and store as `spam-stat'.
447 Every element in ENTRIES has the form \(WORD GOOD BAD) where WORD is
448 the word string, NGOOD is the number of good mails it has appeared in,
449 NBAD is the number of bad mails it has appeared in, GOOD is the number
450 of times it appeared in good mails, and BAD is the number of times it
451 has appeared in bad mails."
452 (let ((table (make-hash-table :size
(length entries
)
456 (spam-stat-make-entry (nth 1 l
) (nth 2 l
))
461 (defun spam-stat-reset ()
462 "Reset `spam-stat' to an empty hash-table.
463 This deletes all the statistics."
465 (setq spam-stat
(make-hash-table :test
'equal
)
468 (setq spam-stat-dirty t
))
472 (defvar spam-stat-score-data nil
473 "Raw data used in the last run of `spam-stat-score-buffer'.")
475 (defsubst spam-stat-score-word
(word)
476 "Return score for WORD.
477 The default score for unknown words is stored in
478 `spam-stat-unknown-word-score'."
479 (spam-stat-score (gethash word spam-stat
)))
481 (defun spam-stat-buffer-words-with-scores ()
482 "Process current buffer, return the 15 most conspicuous words.
483 These are the words whose spam-stat differs the most from 0.5.
484 The list returned contains elements of the form \(WORD SCORE DIFF),
485 where DIFF is the difference between SCORE and 0.5."
486 (let (result word score
)
487 (maphash (lambda (word ignore
)
488 (setq score
(spam-stat-score-word word
)
489 result
(cons (list word score
(abs (- score
0.5)))
491 (spam-stat-buffer-words))
492 (setq result
(sort result
(lambda (a b
) (< (nth 2 b
) (nth 2 a
)))))
493 (setcdr (nthcdr 14 result
) nil
)
497 (defmacro spam-stat-called-interactively-p
(kind)
500 (eval '(called-interactively-p 'any
))
502 `(called-interactively-p ,kind
))
504 (wrong-number-of-arguments '(called-interactively-p))
506 (void-function '(interactive-p)))))
508 (defun spam-stat-score-buffer ()
509 "Return a score describing the spam-probability for this buffer.
510 Add user supplied modifications if supplied."
511 (interactive) ; helps in debugging.
512 (setq spam-stat-score-data
(spam-stat-buffer-words-with-scores))
513 (let* ((probs (mapcar 'cadr spam-stat-score-data
))
514 (prod (apply #'* probs
))
516 (/ prod
(+ prod
(apply #'* (mapcar #'(lambda (x) (- 1 x
))
520 spam-stat-error-holder
521 (spam-stat-score-buffer-user score0
)
524 (if score1s
(+ score0 score1s
) score0
)))
525 (when (spam-stat-called-interactively-p 'any
)
529 (defun spam-stat-score-buffer-user (&rest args
)
534 spam-stat-score-buffer-user-functions
)))
535 (if (memq nil scores
) nil
536 (apply #'+ scores
))))
538 (defun spam-stat-split-fancy ()
539 "Return the name of the spam group if the current mail is spam.
540 Use this function on `nnmail-split-fancy'. If you are interested in
541 the raw data used for the last run of `spam-stat-score-buffer',
542 check the variable `spam-stat-score-data'."
543 (condition-case spam-stat-error-holder
545 (set-buffer spam-stat-buffer
)
546 (goto-char (point-min))
547 (when (> (spam-stat-score-buffer) spam-stat-split-fancy-spam-threshold
)
548 (when (boundp 'nnmail-split-trace
)
549 (mapc (lambda (entry)
550 (push entry nnmail-split-trace
))
551 spam-stat-score-data
))
552 spam-stat-split-fancy-spam-group
))
553 (error (message "Error in spam-stat-split-fancy: %S" spam-stat-error-holder
)
558 (defun spam-stat-strip-xref ()
559 "Strip the Xref header."
561 (mail-narrow-to-head)
562 (when (re-search-forward "^Xref:.*\n" nil t
)
563 (delete-region (match-beginning 0) (match-end 0)))))
565 (autoload 'time-to-number-of-days
"time-date")
567 (defun spam-stat-process-directory (dir func
)
568 "Process all the regular files in directory DIR using function FUNC."
569 (let* ((files (directory-files dir t
"^[^.]"))
570 (max (/ (length files
) 100.0))
574 (when (and (file-readable-p f
)
576 (> (nth 7 (file-attributes f
)) 0)
577 (< (time-to-number-of-days (time-since (nth 5 (file-attributes f
))))
578 spam-stat-process-directory-age
))
579 (setq count
(1+ count
))
580 (message "Reading %s: %.2f%%" dir
(/ count max
))
581 (insert-file-contents-literally f
)
582 (spam-stat-strip-xref)
586 (defun spam-stat-process-spam-directory (dir)
587 "Process all the regular files in directory DIR as spam."
589 (spam-stat-process-directory dir
'spam-stat-buffer-is-spam
))
591 (defun spam-stat-process-non-spam-directory (dir)
592 "Process all the regular files in directory DIR as non-spam."
594 (spam-stat-process-directory dir
'spam-stat-buffer-is-non-spam
))
596 (defun spam-stat-count ()
597 "Return size of `spam-stat'."
599 (hash-table-count spam-stat
))
601 (defun spam-stat-test-directory (dir &optional verbose
)
602 "Test all the regular files in directory DIR for spam.
603 If the result is 1.0, then all files are considered spam.
604 If the result is 0.0, non of the files is considered spam.
605 You can use this to determine error rates.
607 If VERBOSE is non-nil display names of files detected as spam or
608 non-spam in a temporary buffer. If it is the symbol `ham',
609 display non-spam files; otherwise display spam files."
610 (interactive "DDirectory: ")
611 (let* ((files (directory-files dir t
"^[^.]"))
614 (total (length files
))
616 (max (/ total
100.0)); float
620 (when (and (file-readable-p f
)
622 (> (nth 7 (file-attributes f
)) 0))
623 (setq count
(1+ count
))
624 (message "Reading %.2f%%, score %.2f"
625 (/ count max
) (/ score count
))
626 (insert-file-contents-literally f
)
627 (setq buffer-score
(spam-stat-score-buffer))
628 (when (> buffer-score
0.9)
629 (setq score
(1+ score
)))
631 (if (> buffer-score
0.9)
632 (unless (eq verbose
'ham
) (push f display-files
))
633 (when (eq verbose
'ham
) (push f display-files
))))
636 (with-output-to-temp-buffer "*spam-stat results*"
637 (dolist (file display-files
)
640 (message "Final score: %d / %d = %f" score total
(/ score total
))))
642 ;; Shrinking the dictionary
644 (defun spam-stat-reduce-size (&optional count
)
645 "Reduce the size of `spam-stat'.
646 This removes all words that occur less than COUNT from the dictionary.
649 (setq count
(or count
5))
650 (maphash (lambda (key entry
)
651 (when (< (+ (spam-stat-good entry
)
652 (spam-stat-bad entry
))
654 (remhash key spam-stat
)))
656 (setq spam-stat-dirty t
))
658 (defun spam-stat-install-hooks-function ()
659 "Install the spam-stat function hooks."
661 (add-hook 'nnmail-prepare-incoming-message-hook
662 'spam-stat-store-current-buffer
)
663 (add-hook 'gnus-select-article-hook
664 'spam-stat-store-gnus-article-buffer
))
666 (defun spam-stat-unload-hook ()
667 "Uninstall the spam-stat function hooks."
669 (remove-hook 'nnmail-prepare-incoming-message-hook
670 'spam-stat-store-current-buffer
)
671 (remove-hook 'gnus-select-article-hook
672 'spam-stat-store-gnus-article-buffer
))
674 (add-hook 'spam-stat-unload-hook
'spam-stat-unload-hook
)
678 ;;; spam-stat.el ends here