File size: 9,859 Bytes
a561338 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 |
import re
from typing import List, Optional, Tuple, Set
import nltk
from nltk.tokenize import sent_tokenize, word_tokenize
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
from rapidfuzz.distance import Levenshtein
from app.config import (
MIN_WORDS_PER_SENTENCE,
MIN_SENTENCE_LENGTH,
SEQUENCE_THRESHOLD,
EXACT_MATCH_SCORE,
)
nltk.download("punkt")
nltk.download("stopwords")
nltk.download("wordnet")
nltk.download('punkt_tab')
lemmatizer = WordNetLemmatizer()
def normalize_text(text: str) -> str:
if not text:
return ""
text = text.lower()
text = text.replace("\u00ad", "")
text = re.sub(r"-\s*\n\s*", "", text)
text = text.replace("\n", " ")
text = re.sub(r"[^\x20-\x7E]+", " ", text)
text = (text.replace("'", "'").replace("'", "'")
.replace(""", '"').replace(""", '"')
.replace("β", "-").replace("β", "-"))
text = re.sub(r"[^\w\s-]", " ", text)
text = re.sub(r"\s+", " ", text).strip()
tokens = word_tokenize(text)
lemmas = [lemmatizer.lemmatize(tok) for tok in tokens]
normalized = " ".join(lemmas)
normalized = re.sub(r"\s+", " ", normalized).strip()
return normalized
def get_meaningful_sentences(text: str) -> List[str]:
"""Extract meaningful sentences from text."""
sentences = sent_tokenize(text or "")
filtered = []
for s in sentences:
words = word_tokenize(s)
if len(words) >= MIN_WORDS_PER_SENTENCE and len(s.strip()) >= MIN_SENTENCE_LENGTH:
filtered.append(s.strip())
return filtered
def extract_keywords(text: str, max_keywords: int = 5) -> List[str]:
"""Extract top keywords from text for search queries."""
words = word_tokenize((text or "").lower())
stop_words = set(stopwords.words("english"))
filtered = [w for w in words if w.isalpha() and w not in stop_words and len(w) > 3]
freq = nltk.FreqDist(filtered)
return [word for word, _ in freq.most_common(max_keywords)]
def _word_shingles(norm_text: str, k: int = 7) -> List[str]:
tokens = norm_text.split()
if len(tokens) < k:
return []
return [" ".join(tokens[i:i+k]) for i in range(len(tokens) - k + 1)]
def _shingle_sets(a_norm: str, b_norm: str, k: int = 7) -> Tuple[Set[str], Set[str]]:
return set(_word_shingles(a_norm, k)), set(_word_shingles(b_norm, k))
def _jaccard(a: Set[str], b: Set[str]) -> float:
if not a and not b:
return 0.0
inter = len(a & b)
union = len(a | b) or 1
return inter / union
def _containment(a: Set[str], b: Set[str]) -> float:
if not a:
return 0.0
inter = len(a & b)
return inter / len(a)
def _winnowing_hashes(norm_text: str, k: int = 7, w: int = 4) -> List[Tuple[int, int]]:
tokens = norm_text.split()
if len(tokens) < k:
return []
shingles = [" ".join(tokens[i:i+k]) for i in range(len(tokens) - k + 1)]
hashes = [(hash(s) & 0xFFFFFFFF, i) for i, s in enumerate(shingles)]
if w <= 1 or len(hashes) <= w:
return list(dict.fromkeys(hashes))
fps: List[Tuple[int, int]] = []
last_min_abs = -1
for i in range(0, len(hashes) - w + 1):
window = hashes[i:i+w]
min_hash, min_idx = None, None
for j, (h, _) in enumerate(window):
if (min_hash is None) or (h < min_hash) or (h == min_hash and j > (min_idx or -1)):
min_hash, min_idx = h, j
abs_idx = i + (min_idx or 0)
if abs_idx != last_min_abs:
fps.append(hashes[abs_idx])
last_min_abs = abs_idx
return list(dict.fromkeys(fps))
def _winnowing_overlap(a_fp: List[Tuple[int, int]], b_fp: List[Tuple[int, int]]) -> float:
a_set, b_set = set(a_fp), set(b_fp)
if not a_set or not b_set:
return 0.0
shared = len(a_set & b_set)
denom = min(len(a_set), len(b_set)) or 1
return shared / denom
def _exact_substring(norm_sentence: str, norm_external: str) -> bool:
if not norm_sentence or not norm_external:
return False
return norm_external.find(norm_sentence) != -1
def _levenshtein_sim(a: str, b: str) -> float:
if not a or not b:
return 0.0
return Levenshtein.normalized_similarity(a, b)
def _lcs_length(a: str, b: str) -> int:
if not a or not b:
return 0
prev = [0] * (len(b) + 1)
best = 0
for i in range(1, len(a) + 1):
curr = [0]
ai = a[i-1]
for j in range(1, len(b) + 1):
if ai == b[j-1]:
v = prev[j-1] + 1
curr.append(v)
if v > best:
best = v
else:
curr.append(0)
prev = curr
return best
def _extract_phrases(norm_text: str, min_words: int = 3, max_words: int = 7) -> List[str]:
"""Extract all phrases of varying lengths for partial match detection."""
tokens = norm_text.split()
phrases = []
for k in range(min_words, min(max_words + 1, len(tokens) + 1)):
for i in range(len(tokens) - k + 1):
phrases.append(" ".join(tokens[i:i+k]))
return phrases
def _phrase_containment_match(norm_sentence: str, norm_external: str) -> Optional[Tuple[str, float]]:
"""Check if ANY phrase (3-7 words) from sentence appears in external text."""
phrases = _extract_phrases(norm_sentence, min_words=3, max_words=7)
best_phrase = None
best_score = 0.0
for phrase in phrases:
if phrase in norm_external:
score = len(phrase.split()) / len(norm_sentence.split())
if score > best_score:
best_score = score
best_phrase = phrase
if best_phrase and best_score >= 0.4:
return best_phrase, round(best_score, 3)
return None
def find_exact_matches(sentence: str, external_text: str) -> Optional[float]:
"""
Full-sentence lexical match with multiple strategies:
1) Exact substring match β EXACT_MATCH_SCORE (100%)
2) Winnowing fingerprints β robust plagiarism detection
3) Edit-distance (Levenshtein) β catches paraphrased content
4) LCS (longest common substring) β catches missed overlaps
"""
norm_s = normalize_text(sentence)
norm_e = normalize_text(external_text)
if len(norm_s) < MIN_SENTENCE_LENGTH:
return None
# 1) Exact full-sentence match
if _exact_substring(norm_s, norm_e):
return EXACT_MATCH_SCORE
# 2) Winnowing overlap
win_sim = _winnowing_overlap(
_winnowing_hashes(norm_s, k=5, w=4),
_winnowing_hashes(norm_e, k=5, w=4),
)
if win_sim >= SEQUENCE_THRESHOLD * 0.9:
return round(win_sim, 3)
# 3) Edit-distance fallback
lev = _levenshtein_sim(norm_s, norm_e)
if lev >= SEQUENCE_THRESHOLD:
return round(lev, 3)
# 4) LCS fallback
lcs_len = _lcs_length(norm_s, norm_e)
LCS_MIN_CHARS = 15
if lcs_len >= LCS_MIN_CHARS:
return round(lcs_len / max(1, len(norm_s)), 3)
return None
def find_partial_phrase_match(sentence: str, external_text: str) -> Optional[Tuple[str, float]]:
"""
Partial reuse detection via multiple strategies:
1) Phrase containment (3-7 word phrases)
2) Shingle-based Jaccard/Containment
3) Edit distance as fallback
"""
norm_s = normalize_text(sentence)
norm_e = normalize_text(external_text)
if not norm_s or not norm_e:
return None
# Strategy 1: Check if any 3-7 word phrase appears verbatim
phrase_match = _phrase_containment_match(norm_s, norm_e)
if phrase_match:
return sentence, phrase_match[1]
# Strategy 2: 7-word shingles
A, B = _shingle_sets(norm_s, norm_e, k=7)
if A and B:
jac = _jaccard(A, B)
con = _containment(A, B)
score = max(jac, con)
if score >= SEQUENCE_THRESHOLD * 0.8:
return sentence, round(score, 3)
# Strategy 3: Smaller shingles (5-word) for more matches
A_small, B_small = _shingle_sets(norm_s, norm_e, k=5)
if A_small and B_small:
jac_small = _jaccard(A_small, B_small)
con_small = _containment(A_small, B_small)
score_small = max(jac_small, con_small)
if score_small >= SEQUENCE_THRESHOLD * 0.75:
return sentence, round(score_small, 3)
# Strategy 4: Edit distance as last resort
lev = _levenshtein_sim(norm_s, norm_e)
if lev >= SEQUENCE_THRESHOLD * 0.85:
return sentence, round(lev, 3)
return None
def find_partial_phrase_match_for_internal(sentence: str, external_text: str) -> Optional[Tuple[str, float]]:
"""
Wrapper around find_partial_phrase_match that extracts the actual matched phrase
that appears in BOTH documents, not just the full sentence from the first doc.
"""
result = find_partial_phrase_match(sentence, external_text)
if not result:
return None
matched_text, score = result
norm_s = normalize_text(sentence)
norm_e = normalize_text(external_text)
if not norm_s or not norm_e:
return result
# Find the longest common substring that appears in both
words_s = norm_s.split()
words_e = norm_e.split()
best_common = ""
best_len = 0
# Try all consecutive word sequences from sentence A
for i in range(len(words_s)):
for j in range(i + 1, len(words_s) + 1):
phrase = ' '.join(words_s[i:j])
# Check if this phrase exists in document B
if phrase in norm_e:
phrase_len = len(phrase)
if phrase_len > best_len:
best_common = phrase
best_len = phrase_len
# If we found a common phrase, return it
if best_common:
return best_common, score
# Fallback: return original result
return matched_text, score |