| import torch
|
| import re
|
| from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| from typing import List, Dict
|
|
|
|
|
| device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| HF_MODEL_NAME = "alaegmr98/multilingual-scam-detector"
|
|
|
| class AIState:
|
| model = None
|
| tokenizer = None
|
|
|
| state = AIState()
|
|
|
|
|
| SCAM_KEYWORDS = {
|
| 'urgency':[
|
| 'urgent', 'immediately', 'asap', 'warning', 'alert', 'attention', 'critical',
|
| '立刻', '紧急', 'عاجل', 'urgente', 'dringend', 'urgente', 'urgent',
|
| 'act now', 'expires', 'deadline', 'limited time'
|
| ],
|
| 'financial':[
|
| 'bank', 'account', 'credit card', 'paypal', 'wire transfer', 'western union',
|
| '银行卡', '账户', '信用卡', 'بنك', 'حساب', 'banco', 'cuenta', 'konto',
|
| 'investment', 'profit', 'million', 'bitcoin', 'crypto'
|
| ],
|
| 'prize':[
|
| 'won', 'winner', 'lottery', 'prize', 'million', 'inheritance', 'gift',
|
| '中奖', '彩票', 'جائزة', 'ganador', 'lotería', 'gewonnen', 'lotterie',
|
| 'congratulations', 'selected', 'award'
|
| ],
|
| 'threat':[
|
| 'suspend', 'block', 'close', 'terminate', 'legal action', 'lawsuit',
|
| '暂停', '关闭', '法律行动', 'معلقة', 'bloqueado', 'gesperrt',
|
| 'arrest', 'police', 'court', 'fine', 'penalty'
|
| ],
|
| 'personal':[
|
| 'ssn', 'social security', 'password', 'pin', 'credit card number', 'cvv',
|
| '密码', '社会安全号码', 'رقم سري', 'contraseña', 'passwort',
|
| 'identity', 'date of birth', 'mother\'s maiden name'
|
| ]
|
| }
|
|
|
| SAFETY_TIPS = {
|
| 'phishing':[
|
| "Never click on suspicious links - they may steal your information",
|
| "Check the sender's email address carefully for misspellings",
|
| "Hover over links to see the real URL before clicking",
|
| "Contact the company directly using official channels (not the message)"
|
| ],
|
| 'lottery':[
|
| "Legitimate lotteries never ask for money to release prizes",
|
| "You cannot win a lottery you didn't enter",
|
| "Never share bank details for 'prize claims'",
|
| "Delete these messages immediately"
|
| ],
|
| 'bank_fraud':[
|
| "Banks never ask for full passwords or PINs via email/text",
|
| "Call your bank using the number on your card (not the message)",
|
| "Enable two-factor authentication on all accounts",
|
| "Check your account directly by logging into the official website"
|
| ],
|
| 'investment':[
|
| "If it sounds too good to be true, it probably is",
|
| "Research investment opportunities thoroughly",
|
| "Never invest based on an unsolicited message",
|
| "Consult with a financial advisor"
|
| ],
|
| 'general':[
|
| "Don't respond to unsolicited messages",
|
| "Never send money to strangers",
|
| "Keep your personal information private",
|
| "Report suspicious messages to authorities",
|
| "Block the sender immediately"
|
| ]
|
| }
|
|
|
|
|
| def load_model():
|
| try:
|
| print("=" * 60)
|
| print(f"Loading model from Hugging Face: {HF_MODEL_NAME}")
|
| print("=" * 60)
|
| print("This may take a few minutes on first run (downloading model)...")
|
|
|
| state.tokenizer = AutoTokenizer.from_pretrained(HF_MODEL_NAME)
|
| state.model = AutoModelForSequenceClassification.from_pretrained(HF_MODEL_NAME)
|
| state.model.to(device)
|
| state.model.eval()
|
|
|
| print("Model loaded successfully from Hugging Face!")
|
| print(f"Using device: {device}")
|
| print(f"Model type: {state.model.config.model_type}")
|
| print("=" * 60)
|
| except Exception as e:
|
| print(f"Error loading model: {e}")
|
| import traceback
|
| traceback.print_exc()
|
|
|
| def extract_entities(text: str) -> Dict:
|
| entities = {'urls': [], 'phones':[], 'emails': []}
|
|
|
| url_pattern = r'https?://[^\s]+|www\.[^\s]+|[a-zA-Z0-9-]+\.(com|org|net|io|gov|edu|co|uk|de|fr|es|it|ru|cn|jp)(?:/[^\s]*)?'
|
| urls = re.findall(url_pattern, text, re.IGNORECASE)
|
| entities['urls'] = [url for url in urls if url]
|
|
|
| phone_pattern = r'[\+\(]?[1-9][0-9 .\-\(\)]{8,}[0-9]'
|
| phones = re.findall(phone_pattern, text)
|
| entities['phones'] =[phone for phone in phones if len(phone) >= 8]
|
|
|
| email_pattern = r'[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}'
|
| emails = re.findall(email_pattern, text)
|
| entities['emails'] = [email for email in emails if email]
|
|
|
| return entities
|
|
|
| def detect_scam_patterns(text: str) -> Dict:
|
| text_lower = text.lower()
|
| patterns = {'urgency': [], 'financial': [], 'prize': [], 'threat': [], 'personal':[]}
|
| for category, keywords in SCAM_KEYWORDS.items():
|
| for keyword in keywords:
|
| if keyword.lower() in text_lower:
|
| patterns[category].append(keyword)
|
| return patterns
|
|
|
| def get_threat_level(confidence: float, patterns: Dict, has_urls: bool, has_phones: bool) -> str:
|
| score = confidence
|
| if has_urls: score += 15
|
| if has_phones: score += 5
|
| if len(patterns['urgency']) > 0: score += 5 * len(patterns['urgency'])
|
| if len(patterns['financial']) > 0: score += 8 * len(patterns['financial'])
|
| if len(patterns['prize']) > 0: score += 10 * len(patterns['prize'])
|
| if len(patterns['threat']) > 0: score += 8 * len(patterns['threat'])
|
| if len(patterns['personal']) > 0: score += 15 * len(patterns['personal'])
|
|
|
| score = min(score, 100)
|
| if score >= 85: return "CRITICAL"
|
| elif score >= 70: return "HIGH"
|
| elif score >= 50: return "MEDIUM"
|
| else: return "LOW"
|
|
|
| def get_scam_category(patterns: Dict, confidence: float) -> List[str]:
|
| categories = []
|
| if len(patterns['financial']) >= 2 and len(patterns['urgency']) >= 1: categories.append("Banking Phishing Scam")
|
| if len(patterns['prize']) >= 1: categories.append("Lottery/Prize Scam")
|
| if len(patterns['personal']) >= 1: categories.append("Identity Theft Attempt")
|
| if len(patterns['threat']) >= 1: categories.append("Intimidation/Extortion Scam")
|
| if confidence >= 90 and len(patterns['financial']) == 0 and len(patterns['prize']) == 0: categories.append("Generic Phishing")
|
| return categories if categories else ["Suspicious Message"]
|
|
|
| def get_safety_recommendations(categories: List[str], has_urls: bool, has_phones: bool) -> Dict:
|
| tips = set()
|
| actions = set()
|
| for category in categories:
|
| if "Banking" in category or "Phishing" in category:
|
| tips.update(SAFETY_TIPS['bank_fraud'])
|
| tips.update(SAFETY_TIPS['phishing'])
|
| elif "Lottery" in category or "Prize" in category:
|
| tips.update(SAFETY_TIPS['lottery'])
|
| elif "Identity" in category:
|
| tips.update(SAFETY_TIPS['bank_fraud'])
|
| tips.update(SAFETY_TIPS['general'])
|
| elif "Intimidation" in category:
|
| tips.update(SAFETY_TIPS['general'])
|
|
|
| if has_urls:
|
| actions.add("Do NOT click any links in this message")
|
| actions.add("Hover over links to verify the real destination")
|
| if has_phones:
|
| actions.add("Do NOT call any phone numbers in the message")
|
| actions.add("Contact companies using official numbers only")
|
|
|
| actions.add("Delete this message immediately")
|
| actions.add("Block the sender")
|
|
|
| if "Identity" in str(categories):
|
| actions.add("Monitor your accounts for suspicious activity")
|
| actions.add("Consider changing your passwords")
|
|
|
| return {
|
| 'tips': list(tips)[:5] if tips else SAFETY_TIPS['general'][:3],
|
| 'actions': list(actions)[:5]
|
| } |
