1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
(***********************************************************************)
(* *)
(* SpamOracle -- a Bayesian spam filter *)
(* *)
(* Xavier Leroy, projet Cristal, INRIA Rocquencourt *)
(* *)
(* Copyright 2002 Institut National de Recherche en Informatique et *)
(* en Automatique. This file is distributed under the terms of the *)
(* GNU Public License version 2, http://www.gnu.org/licenses/gpl.txt *)
(* *)
(***********************************************************************)
(* $Id$ *)
(* Message ranking *)
open Mail
open Database
let word_count_in w res =
let count = ref 0 in
for i = 0 to Array.length res - 1 do
if w = fst res.(i) then incr count
done;
!count
let add_word w p res =
let i = ref 0 in
while !i < Array.length res
&& abs_float (p -. 0.5) <= abs_float(snd res.(!i) -. 0.5)
do
incr i
done;
if !i < Array.length res then begin
for j = Array.length res - 1 downto !i + 1 do
res.(j) <- res.(j - 1)
done;
res.(!i) <- (w, p)
end
let normalize (p : float) low high =
if p > high then high else if p < low then low else p
let cap (p : float) =
if p > 1.0 then 1.0 else p
let word_proba g b num_g num_b =
let g = 2 * g in (* Graham's magic factor to bias in favor of ham *)
let pgood = cap (float g /. float num_g)
and pbad = cap (float b /. float num_b) in
let p = pbad /. (pgood +. pbad) in
if !Config.robinson_s = 0.0 then
normalize p !Config.low_freq_limit !Config.high_freq_limit
else begin
(* Robinson's adjustement *)
let n = float (g + b) in
let p =
(!Config.robinson_s *. !Config.robinson_x +. n *. p)
/. (!Config.robinson_s +. n) in
(* Result normalization *)
normalize p !Config.low_freq_limit !Config.high_freq_limit
end
let process_word (db, res) w =
try
let (g, b) = Hashtbl.find db.s_freq w in
if word_count_in w res < !Config.max_repetitions then begin
let p = word_proba g b db.s_num_good db.s_num_spam in
add_word w p res
end
with Not_found ->
()
let process_words ((db, res) as ctx) txt =
Wordsplit.iter (process_word ctx) (in_short db) txt;
if !Config.summarize_referenced then Refhosts.add txt
let process_msg ctx m =
iter_message (process_words ctx) m
(* This is Graham's original approach *)
let spaminess_score_graham res =
let p = ref 1.0 and pexp = ref 0
and cp = ref 1.0 and cpexp = ref 0 in
for i = 0 to Array.length res - 1 do
let (_, x) = res.(i) in
p := !p *. x;
if !p <= 1e-100 then begin
let (m, e) = frexp !p in p := m; pexp := !pexp + e
end;
cp := !cp *. (1.0 -. x);
if !cp <= 1e-100 then begin
let (m, e) = frexp !cp in cp := m; cpexp := !cpexp + e
end
done;
if !cpexp < !pexp then cp := ldexp !cp (!cpexp - !pexp)
else if !cpexp > !pexp then p := ldexp !p (!pexp - !cpexp);
!p /. (!p +. !cp)
(* This is Robinson's chi-square stuff *)
let chi2_inverse m n = (* chi2 inverse of 2m with 2n degrees *)
let t = ref (exp (-. m)) in
let s = ref !t in
for i = 1 to n do
t := !t *. m /. float i;
s := !s +. !t
done;
if !s >= 1.0 then 1.0 else !s
let log2 = log 2.0
let chi2_hypothesis ps =
(* Compute -2 * ln (product ps). Be careful with underflows. *)
let p = ref 1.0 and pexp = ref 0 in
for i = 0 to Array.length ps - 1 do
p := !p *. ps.(i);
if !p <= 1e-100 then begin
let (x, e) = frexp !p in p := x; pexp := !pexp + e
end
done;
chi2_inverse (-. (log !p +. log2 *. float !pexp)) (Array.length ps)
let spaminess_score_robinson res =
let probs = Array.map snd res in
let cprobs = Array.map (fun x -> 1.0 -. x) probs in
0.5 *. (1.0 +. chi2_hypothesis probs -. chi2_hypothesis cprobs)
type rank =
{ spam_prob: float;
num_meaningful: int;
explanation: string }
let rank_message db msg =
Refhosts.reset();
let res = Array.make !Config.num_words_retained ("", 0.5) in
process_msg (db, res) msg;
let p =
if !Config.use_chi_square
then spaminess_score_robinson res
else spaminess_score_graham res in
let meaningful = ref 0 in
while !meaningful < Array.length res && fst res.(!meaningful) <> ""
do incr meaningful done;
let summary = Buffer.create 200 in
for i = 0 to !meaningful - 1 do
let (w, p) = res.(i) in
Printf.bprintf summary "%s:%02d " w (truncate (p *. 100.0))
done;
{ spam_prob = p;
num_meaningful = !meaningful;
explanation = Buffer.contents summary }
|