udbhav89/domain-specific-modern-bert-spam-classifier

Domain-Specific ModernBERT Spam Classifier

Fine-tuned ModernBERT model for spam classification with domain-specific prefixes.

Usage

This model expects text with a domain prefix:

• —[crypto] for crypto/blockchain related content
• —[rideshare] for rideshare/transportation related content

Example:

from transformers import pipeline

classifier = pipeline("text-classification", model="udbhav89/domain-specific-modern-bert-spam-classifier")

# Crypto domain
result = classifier("[crypto] FREE BITCOIN! Send 0.1 ETH to get 10 BTC back!")
# Returns: [{'label': 'spam', 'score': 0.97}]

# Rideshare domain
result = classifier("[rideshare] Great ride, thanks driver!")
# Returns: [{'label': 'not_spam', 'score': 0.99}]

Labels

• —not_spam (0): Legitimate message
• —spam (1): Spam message

Training

• —Base model: answerdotai/ModernBERT-base
• —Training data: 50k samples (25k crypto, 25k rideshare)
• —Focal loss for imbalanced data
• —Trained on A100 GPU