Moustapha2002-git/hassaniya-sentiment

Sentiment Analysis for Hassaniya dialect using MARBERTv2 Fine-tuning # Hassaniya Dialect — Sentiment Analysis Fine-tuning of MARBERTv2 for sentiment classification of the Hassaniya dialect (Mauritania). ## Results | Model | Accuracy | F1-Score | |-------|----------|----------| | MARBERTv2 + Data Augmentation | **97.16%** | **> 0.96** | ## Dataset - 1 851 original phrases → 7 404 after augmentation ×4 - 3 classes : Positive / Neutral / Negative - Source: Mendeley Data — Université de Nouakchott ## Method - Preprocessing + 62 Hassaniya stopwords - Data Augmentation ×4 (deletion, swap, insertion) - Fine-tuning MARBERTv2 (10 epochs + Early Stopping) - Class weights to balance Neutral class - Interactive demo with Gradio (Accuracy: 97.16%) ## Files | File | Description | |------|-------------| | `Hassaniya_Sentiment_Analysis_Final.ipynb` | Main notebook | | `projectHA_DATASET_-_VF.csv` | Hassaniya dataset (1 851 phrases) | | `H-Stopwords.txt` | Hassaniya stopwords list (62 words) | ## Author **Moustapha Yahdih** Master 1 Artificial Intelligence — NLP & Computer Vision Université de Nouakchott — 2025/2026