-- Different text parts
rspamd_config.R_PARTS_DIFFER = function(task)
- local distance = task:get_mempool():get_variable('parts_distance', 'int')
+ local distance = task:get_mempool():get_variable('parts_distance', 'double')
if distance then
local nd = tonumber(distance)
-
- if nd < 50 then
- local score = 1 - util.tanh(nd / 100.0)
-
- task:insert_result('R_PARTS_DIFFER', score, tostring(nd) .. '%')
+ -- ND is relation of different words to total words
+ if nd >= 0.5 then
+ local tw = task:get_mempool():get_variable('total_words', 'int')
+
+ if tw then
+ if tw > 30 then
+ -- We are confident about difference
+ local score = (nd - 0.5) * 2.0
+ else
+ -- We are not so confident about difference
+ local score = (nd - 0.5)
+ end
+ task:insert_result('R_PARTS_DIFFER', score, tostring(100.0 * nd) .. '%')
+ end
end
end
const gchar *p;
gsize len;
goffset hdr_pos;
- gint diff, *pdiff, i;
- guint tw, dw;
+ gint i;
+ gdouble diff, *pdiff;
+ guint tw, *ptw, dw;
tmp = rspamd_mempool_alloc (task->task_pool, sizeof (GByteArray));
p = task->msg.begin;
if (!IS_PART_EMPTY (p1) && !IS_PART_EMPTY (p2) &&
p1->normalized_words && p2->normalized_words) {
- tw = MAX (p1->normalized_words->len, p2->normalized_words->len);
+ tw = p1->normalized_words->len + p2->normalized_words->len;
if (tw > 0) {
dw = rspamd_words_levenshtein_distance (task,
p1->normalized_words,
p2->normalized_words);
- diff = (100.0 * (gdouble)(tw - dw) / (gdouble)tw);
+ diff = (2.0 * (gdouble)dw) / (gdouble)tw;
- debug_task (
+ msg_err_task (
"different words: %d, total words: %d, "
- "got likeliness between parts of %d%%",
+ "got diff between parts of %.2f",
dw, tw,
diff);
- pdiff = rspamd_mempool_alloc (task->task_pool, sizeof (gint));
+ pdiff = rspamd_mempool_alloc (task->task_pool,
+ sizeof (gdouble));
*pdiff = diff;
rspamd_mempool_set_variable (task->task_pool,
"parts_distance",
pdiff,
NULL);
+ ptw = rspamd_mempool_alloc (task->task_pool,
+ sizeof (gint));
+ *ptw = tw;
+ rspamd_mempool_set_variable (task->task_pool,
+ "total_words",
+ ptw,
+ NULL);
}
}
}
gboolean
rspamd_parts_distance (struct rspamd_task * task, GArray * args, void *unused)
{
- gint threshold, threshold2 = -1, diff;
+ gint threshold, threshold2 = -1;
struct expression_argument *arg;
- gint *pdiff;
+ gdouble *pdiff, diff;
if (args == NULL || args->len == 0) {
debug_task ("no threshold is specified, assume it 100");
if ((pdiff =
rspamd_mempool_get_variable (task->task_pool,
"parts_distance")) != NULL) {
- diff = *pdiff;
+ diff = (1.0 - (*pdiff)) * 100.0;
+
if (diff != -1) {
if (threshold2 > 0) {
- if (diff >=
- MIN (threshold,
- threshold2) && diff < MAX (threshold, threshold2)) {
+ if (diff >= MIN (threshold, threshold2) &&
+ diff < MAX (threshold, threshold2)) {
+
return TRUE;
}
}
#define RSPAMD_LEARN_OP 1
#define RSPAMD_UNLEARN_OP 2
-static const gint similarity_treshold = 80;
+static const gdouble similarity_treshold = 80.0;
static void
rspamd_stat_tokenize_header (struct rspamd_task *task,
GArray *words;
gchar *sub;
guint i, reserved_len = 0;
- gint *pdiff;
+ gdouble *pdiff;
for (i = 0; i < task->text_parts->len; i++) {
part = g_ptr_array_index (task->text_parts, i);
}
- if (pdiff != NULL && *pdiff > similarity_treshold) {
+ if (pdiff != NULL && (1.0 - *pdiff) * 100.0 > similarity_treshold) {
msg_debug_task ("message has two common parts (%d%%), so skip the last one",
*pdiff);
break;