Moderation improvements (#564)

banned-club.sql

@@ -800,6 +800,7 @@ VALUES
     ('fruits as a metaphor for sex'),
     ('fuck'),
     ('fuck 12'),
+    ('fuckable'),
     ('fuck acab'),
     ('fuckaduck'),
     ('fuck any colors'),
@@ -2287,6 +2288,7 @@ VALUES
     ('rag heads'),
     ('ragheads'),
     ('rape'),
+    ('rapeable'),
     ('rapebait'),
     ('rapebastard club'),
     ('rapebastardclub'),

chat.requirements.txt

@@ -1,3 +1,4 @@
+confusable-homoglyphs
 lxml
 nltk
 psycopg[binary]

service/chat/__init__.py

@@ -33,7 +33,7 @@
 
 PORT = sys.argv[1] if len(sys.argv) >= 2 else 5443
 
-MAX_INTROS_PER_DAY = 50
+MAX_INTROS_PER_DAY = 35
 
 # TODO: Tables to migrate to monolithic DB:
 #

service/chat/offensive/__init__.py

@@ -1,3 +1,4 @@
+from confusable_homoglyphs import confusables
 import re
 import unicodedata
 
@@ -36,6 +37,8 @@
     "fvck": "fuck",
     "fvcked": "fucked",
     "fvcking": "fucking",
+    "ngger": "nigger",
+    "nggr": "nigger",
     "nggr": "nigger",
     "p0rn": "porn",
     "pissin": "pissing",
@@ -54,6 +57,7 @@
     "anal",
     "anally",
     "anus",
+    "around your throat",
     "ass fuck",
     "ass fucked",
     "ass fucker",
@@ -66,7 +70,10 @@
     "assfucked",
     "assfucker",
     "assfucking",
+    "back shots",
+    "backshots",
     "ballsack",
+    "bash your",
     "beastial",
     "beastiality",
     "bellend",
@@ -91,6 +98,7 @@
     "butt-hole",
     "butthole",
     "buttstuff",
+    "bwc",
     "carpet muncher",
     "cawk",
     "cervix",
@@ -131,13 +139,27 @@
     "cuntlicking",
     "cunts",
     "cut me",
+    "cut my wrist",
+    "cut my wrists",
+    "cut myself",
     "cut you",
+    "cut your wrist",
+    "cut your wrists",
+    "cut yourself",
+    "cutting myself",
+    "cutting yourself",
+    "deep throat",
+    "deep throated",
+    "deepthroat",
     "dick",
     "dildo",
     "dildos",
+    "do it raw",
     "dog-fucker",
     "doggin",
     "dogging",
+    "down my throat",
+    "down your throat",
     "dyke",
     "dykes",
     "e sex",
@@ -154,6 +176,7 @@
     "ejaculatings",
     "ejaculation",
     "ejakulate",
+    "end your life",
     "esex",
     "faggitt",
     "faggot",
@@ -186,8 +209,10 @@
     "fuck your face",
     "fuck your mouth",
     "fuck your thighs",
+    "fuck your throat",
     "fuck your tits",
     "fuck yourself",
+    "fuckable",
     "fucking me",
     "fucking you",
     "gag me",
@@ -214,6 +239,7 @@
     "hanging myself",
     "hanging yourself",
     "heil",
+    "hit it raw",
     "hoe",
     "hoes",
     "horniest",
@@ -222,6 +248,8 @@
     "incest",
     "jack-off",
     "jackoff",
+    "jeet",
+    "jeets",
     "jerk off",
     "jerk-off",
     "jerked off",
@@ -245,6 +273,8 @@
     "labia",
     "lick my",
     "lick your",
+    "like it raw",
+    "likes it raw",
     "loli",
     "lolicon",
     "masterbate",
@@ -259,6 +289,7 @@
     "molester",
     "molesting",
     "my nuts",
+    "my throat",
     "necrophilia",
     "nigga",
     "niggas",
@@ -273,6 +304,12 @@
     "orgy",
     "paedo",
     "paedophile",
+    "pajeet",
+    "pajeeta",
+    "pajeetas",
+    "pajeets",
+    "paki",
+    "pakis",
     "paraphilias",
     "pedo",
     "pedophile",
@@ -281,9 +318,15 @@
     "penis",
     "penisfucker",
     "phonesex",
+    "pin you",
+    "piss",
     "pissflaps",
     "pissin",
     "pissing",
+    "poojeet",
+    "poojeeta",
+    "poojeetas",
+    "poojeets",
     "porn",
     "porno",
     "pornography",
@@ -294,11 +337,13 @@
     "pussys",
     "rail you",
     "rape",
+    "rapeable",
     "rapebait"
     "raped",
     "rapes",
     "raping",
     "rapist",
+    "raw dog you",
     "retard",
     "retardation",
     "retarded",
@@ -319,20 +364,40 @@
     "shota",
     "shotacon",
     "skank",
+    "slit my wrist",
+    "slit my wrists",
+    "slit your wrist",
+    "slit your wrists",
     "slut",
     "sluts",
+    "slutty",
     "smegma",
     "sodomize",
     "sodomy",
     "some head",
     "spic",
+    "spit in my face",
+    "spit in my mouth",
+    "spit in your face",
+    "spit in your mouth",
+    "spit on me",
+    "spit on my face",
+    "spit on my mouth",
+    "spit on you",
+    "spit on your face",
+    "spit on your mouth",
     "stabbing me",
     "stabbing you",
     "strangle me",
     "strangle you",
     "suicidal",
     "suicide",
     "testicle",
+    "throat fuck",
+    "throat fucking",
+    "throat pussy",
+    "throatfuck",
+    "throatfucking",
     "tie me",
     "tie you",
     "tit fuck",
@@ -376,9 +441,13 @@
     "whore",
     "whores",
     "wincest",
+    "you are retarded",
+    "you retard",
+    "you retarded",
     "you will never be a woman",
     "you'll never be a woman",
     "your throat",
+    "your throat",
     "ywnbaw",
     "zoophilia",
 ]
@@ -390,17 +459,47 @@
 
 
 # Characters which were repeated more than once
-_repeated_characters_pattern = re.compile(r'(.)\1+')
+_repeated_characters_pattern = re.compile(r'(.)\1+', re.IGNORECASE)
 
 
 _offensive_matcher = re.compile(_offensive_pattern, re.IGNORECASE)
 
 
-def _apply_normalization_map(haystack: str):
+def _get_latin_homoglyph(char: str) -> str:
+    """
+    Returns a Latin homoglyph for the given character if available.
+    If no Latin homoglyph is found, returns the original character.
+    """
+    # confusables.is_confusable returns a list of dictionaries, one per confusable character
+    # Each dictionary can contain a 'homoglyphs' key, which is a list of homoglyph entries.
+    info_seq = confusables.is_confusable(char, preferred_aliases=['latin']) or []
+
+    for info in info_seq:
+        # Extract the first Latin homoglyph character if one exists
+        latin_homoglyphs = (
+                h['c']
+                for h in info.get('homoglyphs', [])
+                if info.get('alias') not in ['LATIN', 'COMMON']
+        )
+        latin_homoglyph = next(latin_homoglyphs, None)
+        if latin_homoglyph:
+            return latin_homoglyph
+
+    return char
+
+
+def _normalize_homoglyphs(s: str) -> str:
+    """
+    Normalizes an input string by replacing characters that are confusable with Latin homoglyphs.
+    """
+    return ''.join(_get_latin_homoglyph(char) for char in s)
+
+
+def _normalize_spelling(haystack: str):
     for needle, replacement in _normalization_map.items():
         # Apparently compiled regexes are cached between invocations of
         # re.compile.
-        pattern = re.compile(f"\\b{needle}\\b")
+        pattern = re.compile(f"\\b{needle}\\b", re.IGNORECASE)
 
         haystack = pattern.sub(replacement, haystack)
 
@@ -415,6 +514,9 @@ def normalize_string(s: str):
         char for char in normalized_input if not unicodedata.combining(char)
     )
 
+    # Mitigate homoglyph attacks
+    normalized_input = _normalize_homoglyphs(normalized_input)
+
     # Normalize whitespace
     normalized_input = ' '.join(_split_pattern.split(normalized_input))
 
@@ -423,7 +525,7 @@ def normalize_string(s: str):
         r'\1\1', normalized_input)
 
     # Replace slang
-    normalized_input = _apply_normalization_map(normalized_input)
+    normalized_input = _normalize_spelling(normalized_input)
 
     return normalized_input
 

service/chat/offensive/test_init.py

@@ -7,6 +7,9 @@ def test_offensive_strings(self):
         self.assertTrue(
                 is_offensive("go fk urself please"))
 
+        self.assertTrue(
+                is_offensive("GO FK URSELF PLEASE"))
+
         self.assertTrue(
                 is_offensive("I want to cvm on u"))
 
@@ -25,6 +28,19 @@ def test_offensive_strings(self):
         self.assertTrue(
                 is_offensive("Would you like to purchase viagra?"))
 
+        # whore is offensive even if followed by an emoji without space between
+        self.assertTrue(
+                is_offensive("you are a whore🙂"))
+
+        self.assertTrue(
+                is_offensive("you are a whore!"))
+
+        # Homoglyph attacks. The letter "o" in "whоre" has been replaced with a
+        # Cyrillic "о" (Unicode character U+043E), which looks identical to the
+        # Latin "o" (Unicode character U+006F)
+        self.assertTrue(
+                is_offensive("you filthy degеnerate pоrn addicted whоre !"))
+
         self.assertTrue(
                 is_offensive("I am the rapist"))
 
@@ -47,6 +63,21 @@ def test_inoffensive_strings(self):
         self.assertFalse(
                 is_offensive("I am therapist"))
 
+    # Using Cyrillic characters
+    def test_cyrillic_substitution(self):
+        # Substituting "a" with Cyrillic "а" (U+0430)
+        self.assertTrue(is_offensive("gаg me"))
+        # Substituting "e" with Cyrillic "е" (U+0435)
+        self.assertTrue(is_offensive("gave mе head"))
+        # Mixed Latin and Cyrillic characters
+        self.assertTrue(is_offensive("fuckаblе"))
+
+    # Using Greek characters
+    def test_greek_substitution(self):
+        # Substituting "o" with Greek "ο" (U+03BF)
+        self.assertTrue(is_offensive("gοok"))
+        # Substituting "i" with Greek "ι" (U+03B9)
+        self.assertTrue(is_offensive("gιmme head"))
 
 
 if __name__ == '__main__':