bethel4/amharic-ecommerce-ner

# amharic-ecommerce-ner # Amharic E-commerce NER This project extracts product, price, location, and other key business entities from Amharic Telegram e-commerce posts. Built for EthioMart's goal to centralize vendor intelligence and improve credit scoring. ## Objectives - Ingest Telegram e-commerce messages (text/images) - Preprocess and structure Amharic data - Label and fine-tune transformer models (mBERT, XLM-R) for NER - Evaluate and explain model predictions (SHAP, LIME) ## Project Structure ``` amharic-ecommerce-ner/ │ ├── data/ │ ├── raw/ # Unprocessed scraped data │ ├── processed/ # Cleaned, structured data │ └── external/ # Any downloaded datasets (e.g., Amharic NER dataset) │ ├── notebooks/ # Jupyter notebooks for EDA, training, etc. │ ├── 01_scraping.ipynb │ ├── 02_preprocessing.ipynb │ ├── 03_labeling.ipynb │ └── 04_model_training.ipynb │ ├── scripts/ # Script versions of notebooks or automation │ ├── scraper.py │ ├── preprocess.py │ └── fine_tune.py │ ├── src/ # Core logic as importable modules │ ├── data_ingestion/ │ ├── preprocessing/ │ ├── labeling/ │ └── modeling/ │ ├── models/ # Saved fine-tuned model files │ ├── outputs/ # Visualizations, results, metrics, SHAP/LIME outputs │ ├── tests/ # Unit tests │ ├── .gitignore ├── README.md ├── requirements.txt └── LICENSE ```