AraBERT for Arabic Sentiment Analysis

Fine-tuned aubmindlab/bert-base-arabertv02 for Arabic sentiment classification (positive/negative).

🎯 Model Description

This model classifies Arabic text into positive or negative sentiment with 92.87% accuracy. Built using transfer learning on BERT-base-arabertv02 with task-specific fine-tuning for binary sentiment analysis.

🚀 Quick Start

from transformers import pipeline

# Load model
classifier = pipeline("sentiment-analysis", model="Belall87/arabert-arabic-sentiment")

# Predict
result = classifier("هذا المنتج رائع جداً")
print(result)
# [{'label': 'POSITIVE', 'score': 0.95}]

# Batch prediction
texts = [
    "الخدمة ممتازة والموظفين متعاونين",
    "تجربة سيئة جداً ولا أنصح بها"
]
results = classifier(texts)

📊 Performance

Metric	Value
Accuracy	92.87%
F1-Score	92.87%
Precision	92.85%
Recall	92.89%

🔧 Training Details

Base Model

Architecture: BERT-base (12 layers, 768 hidden, 12 attention heads)
Pre-trained: aubmindlab/bert-base-arabertv02
Parameters: ~110M

Training

Optimizer: AdamW
Learning Rate: 2e-5 with cosine scheduling
Batch Size: 8
Epochs: 3
Max Length: 256 tokens
Hardware: GPU (fp16 mixed precision)

Data Preprocessing

Arabic text normalization (alef, yeh, hamza variants)
Diacritics removal
URL, mention, hashtag filtering
Stratified train/val/test split (72%/8%/20%)

💡 Intended Use

Direct Use

Arabic social media sentiment monitoring
Product review analysis
Customer feedback classification
Opinion mining in Arabic text

Limitations

Binary classification only (positive/negative)
Trained on Modern Standard Arabic (MSA) and dialectal mix
May underperform on domain-specific jargon
Best with text length < 256 tokens

📚 Training Data

Arabic sentiment dataset with stratified sampling for balanced training. Preprocessing includes normalization of Arabic orthographic variants and removal of noise.

🔗 Links

GitHub Repository: Arabic-Sentiment-Analysis-BiLSTM-vs-AraBERT
Kaggle Notebook: Arabic Sentiment Analysis
Base Model: bert-base-arabertv02

📝 Citation

@misc{belal-arabert-sentiment-2025,
  author = {Belal Mahmoud Hussien},
  title = {AraBERT for Arabic Sentiment Analysis},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Belall87/arabert-arabic-sentiment}}
}

📧 Contact

Belal Mahmoud Hussien

📄 License

MIT License

Downloads last month: 3

Safetensors

Model size

0.1B params

Tensor type

F32