---
license: mit
language:
- ar
metrics:
- f1
base_model:
- UBC-NLP/MARBERTv2
pipeline_tag: text-classification
library_name: transformers
---

# 🥤 SODA-BERT

Fine-tuned Arabic language model based on [UBC-NLP/MARBERTv2](https://huggingface.co/UBC-NLP/MARBERTv2), trained on the **OmanSent** dataset, the first dataset produced using the **SODA** data collection framework. This model focuses on sentiment analysis and text classification tasks in Arabic, with a particular emphasis on Omani and Gulf dialects.

## 📊 Model Details

- **Base model**: [`UBC-NLP/MARBERTv2`](https://huggingface.co/UBC-NLP/MARBERTv2)
- **Fine-tuning dataset**:
  - OmanSent (Omani dialect sentiment dataset, collected using the SODA framework — not yet publicly released)
- **Languages**: Arabic (Modern Standard Arabic + Gulf/Omani dialects)
- **Task**: Sentiment Analysis / Text Classification

## 🛠️ How to Use

```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("mktr/SODA-BERT")
model = AutoModelForSequenceClassification.from_pretrained("mktr/SODA-BERT")

text = "الي يقول العماني ما مال شغل تفل في وجهه"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
predictions = outputs.logits.argmax(dim=-1)

# Map prediction to sentiment label
label_map = {0: "Negative", 1: "Positive", 2: "Neutral"}
predicted_label = label_map[predictions.item()]

print(f"Predicted Sentiment: {predicted_label}")
```