nahiar/hatespeech-xlmr-v4

Hate Speech Detection for Social Media Text

Multilingual Indonesian & English — XLM-RoBERTa

This repository provides a fine-tuned XLM-RoBERTa model for Hate Speech detection in social media text.
The model is designed to identify hate speech vs non-hate speech across Indonesian, regional Indonesian languages, and English, particularly in noisy and informal online conversations.

---

🚀 Highlights

• Binary classification: Hate Speech / Non-Hate
• Multilingual support (Indonesia + English)
• Robust on informal and user-generated content
• Ready-to-use with Hugging Face pipeline
• Suitable for content moderation and safety systems

---

🌍 Supported Languages

• 🇮🇩 Bahasa Indonesia
• Bahasa Melayu
• Indonesian regional languages (Aceh, Banjar, Bugis, Jawa, Madura, Minang, Sunda, dll.)
• 🇬🇧 English

---

📊 Model Performance

Performance metrics are reported on a held-out validation set.

Metric	Score
Accuracy	0.9180
F1 (Macro)	0.9179
F1 (Weighted)	0.9200
Training Loss	0.9200
Validation Loss	0.9200

(Exact scores may vary depending on evaluation split and threshold.)

---

⚙️ Usage

Installation

pip install transformers torch

Single Prediction

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="nahiar/hatespeech-xlmr-v4"
)

result = classifier("Dasar bodoh, otak udang!")
print(result)

Output

[{'label': 'LABEL_1', 'score': 0.9821}]

Label Mapping

LABEL_0 → NON_HATE
LABEL_1 → HATE_SPEECH

---

📦 Batch Inference

texts = [
    "Kamu itu memang tidak berguna",
    "Saya tidak setuju dengan pendapat kamu",
    "Dasar kaum ini selalu bikin rusuh"
]

results = classifier(texts)

for text, result in zip(texts, results):
    print(f"{result['label']} ({result['score']:.4f}) → {text}")

---

🏗️ Training Configuration

Parameter	Value
Base Model	xlm-roberta-base
Training Strategy	Fine-tuning
Epochs	Multiple
Learning Rate	2e-5
Batch Size	16
Training Date	2025-12-03

---

🎯 Intended Use

• Hate speech detection and moderation
• Content safety and compliance systems
• Pre-filtering for sentiment or topic analysis
• Social media monitoring dashboards

---

⚠️ Limitations

• Binary classification only (Hate / Non-Hate)
• Does not identify hate targets or categories
• Context-dependent sarcasm may be misclassified
• Not suitable for legal judgment without human review

---

📜 License

This model is released under the Apache License 2.0 Free for research and commercial use.

---

📚 Citation

@misc{djunaedi2025hatespeech,
  author    = {Raihan Hidayatulloh Djunaedi},
  title     = {Hate Speech Detection for Social Media Text},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/nahiar/hatespeech-xlmr-v4}
}

---

🙌 Acknowledgements

• Hugging Face Transformers
• Facebook AI Research — XLM-RoBERTa