Fikerdeep/Amharic-Sentiment-Analysis

# Amharic Sentiment Analysis A deep learning-based sentiment analysis system for Amharic (Ethiopian) text. This project implements multiple neural network architectures for binary sentiment classification (positive/negative). ## Features - **Multiple Model Architectures**: CNN, BiLSTM, GRU, and hybrid CNN-BiLSTM - **Framework Support**: TensorFlow/Keras, PyTorch, and Hugging Face Transformers - **Amharic-Specific Preprocessing**: Handles Ge'ez script character variants and labialized characters - **Production-Ready API**: FastAPI REST API with Docker support - **Comprehensive Evaluation**: Accuracy, precision, recall, F1-score, ROC-AUC metrics ## Model Performance | Model | Accuracy | Precision | Recall | F1 Score | |-------|----------|-----------|--------|----------| | CNN | 84.8% | 80.4% | 73.7% | - | | GRU | 88.6% | 88.0% | 91.5% | - | | BiLSTM | 87.6% | 84.2% | 92.9% | - | | **CNN-BiLSTM** | **91.6%** | **90.5%** | **93.9%** | - | ## Project Structure ``` Amharic-Sentiment-Analysis/ ├── amharic_sentiment/ # Main package │ ├── preprocessing/ # Text cleaning and normalization │ ├── data/ # Dataset and data loading utilities │ ├── models/ # TensorFlow/Keras models │ ├── pytorch/ # PyTorch models │ ├── transformers/ # Hugging Face Transformers │ ├── training/ # Training pipeline │ ├── evaluation/ # Metrics and visualization │ └── utils/ # Configuration and logging ├── api/ # FastAPI REST API ├── docker/ # Docker configuration ├── configs/ # YAML configuration files ├── scripts/ # Utility scripts ├── notebooks/ # Jupyter notebooks └── dataset/ # Training data ``` ## Installation ### From Source ```bash # Clone the repository git clone github.com …