Arabic ABSA Model

📌 Overview

This project implements an Aspect-Based Sentiment Analysis (ABSA) model for Arabic customer feedback.

The model predicts sentiment (positive, negative, neutral) for predefined aspects in each review.

---

📂 Project Structure

• train_rewritten.ipynb → Training + inference code
• submission.json → Final predictions
• arabert-absa/ → Saved model weights
• requirements.txt → Dependencies

---

⚙️ Setup

Install dependencies:

pip install -r requirements.txt

---

🚀 How to Run

1. Train Model

Run all cells in the notebook until:

trainer.train()

---

2. Run Inference (Prediction)

Run the prediction section to generate:

DeepX_submission.json

---

🧠 Model Details

• Base model: AraBERT

• Task: Aspect-Based Sentiment Analysis

• Input format:

  review_text [SEP] aspect
  

---

📊 Output Format

[
  {
    "review_id": 1,
    "aspects": ["food", "service"],
    "aspect_sentiments": {
      "food": "positive",
      "service": "negative"
    }
  }
]

---

⚠️ Notes

• Datasets are NOT included as per competition rules
• Model weights are included in /arabert-absa/

---

👩‍💻 Team

Yomna Saad

Dalia Nassar

Maryam Mahmoud