Indonesian Spam Detection Model

Model Overview

Indonesian Spam Detection Model is a fine-tuned spam detection model based on the Gemma 2 2B architecture. This model is specifically designed for identifying spam messages in Indonesian text, particularly for WhatsApp chatbot interactions. It has been fine-tuned using a comprehensive dataset of 40,000 spam messages collected over a year.

Labels

The model classifies text into two categories:

0: Non-spam (legitimate message)
1: Spam (unwanted/malicious message)

Detection Capabilities

The model can effectively detect various types of spam including:

Offensive and abusive language
Profane content
Gibberish text and random characters
Suspicious links and URLs
Promotional spam
Fraudulent messages

Use this Model

Installation

First, install the required dependencies:

pip install transformers torch

Quick Start

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load model and tokenizer
model_name = "nahiar/spam-analysis"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Example texts to classify
texts = [
    "Halo, bagaimana kabar Anda hari ini?",  # Non-spam
    "MENANG JUTAAN RUPIAH! Klik link ini sekarang: http://suspicious-link.com",  # Spam
    "adsfwcasdfad12345",  # Spam (gibberish)
    "Terima kasih atas informasinya"  # Non-spam
]

# Tokenize and predict
for text in texts:
    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)

    with torch.no_grad():
        outputs = model(**inputs)
        prediction = torch.nn.functional.softmax(outputs.logits, dim=-1)
        predicted_class = torch.argmax(prediction, dim=1).item()
        confidence = torch.max(prediction, dim=1)[0].item()

    label = "Spam" if predicted_class == 1 else "Non-spam"
    print(f"Text: {text}")
    print(f"Prediction: {label} (confidence: {confidence:.4f})")
    print("-" * 50)

Batch Processing

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

def classify_spam_batch(texts, model_name="nahiar/spam-analysis"):
    """
    Classify multiple texts for spam detection

    Args:
        texts (list): List of texts to classify
        model_name (str): Hugging Face model name

    Returns:
        list: List of predictions with confidence scores
    """
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSequenceClassification.from_pretrained(model_name)

    # Tokenize all texts
    inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True, max_length=512)

    with torch.no_grad():
        outputs = model(**inputs)
        predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
        predicted_classes = torch.argmax(predictions, dim=1)
        confidences = torch.max(predictions, dim=1)[0]

    results = []
    for i, text in enumerate(texts):
        results.append({
            'text': text,
            'is_spam': bool(predicted_classes[i].item()),
            'confidence': confidences[i].item(),
            'label': 'Spam' if predicted_classes[i].item() == 1 else 'Non-spam'
        })

    return results

# Example usage
texts = [
    "Selamat pagi, semoga harimu menyenangkan",
    "URGENT!!! Dapatkan uang 10 juta hanya dengan klik link ini",
    "Terima kasih sudah membantu kemarin"
]

results = classify_spam_batch(texts)
for result in results:
    print(f"Text: {result['text']}")
    print(f"Label: {result['label']} (Confidence: {result['confidence']:.4f})")
    print()

Model Performance

This model has been trained on a diverse dataset of Indonesian text messages and demonstrates strong performance in distinguishing between spam and legitimate messages across various contexts including:

WhatsApp chatbot interactions
SMS messages
Social media content
Customer service communications

Limitations

The model is primarily trained on Indonesian language text
Performance may vary with very short messages (< 10 characters)
Context-dependent spam (messages that are spam only in specific contexts) may be challenging

Repository

For more information about the training process and code implementation, visit:

https://github.com/nahiar/spam-analysis

Citation

If you use this model in your research or applications, please cite:

@misc{spam-analysis-indo,
  title={Indonesian Spam Detection Model},
  author={Nahiar},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/nahiar/spam-analysis}
}

nahiar
/

spam-analysis