EstiphanosH/Amharic-Ecommerce-Data-Extractor

# 🛍️ Amharic E-commerce Data Extractor A Telegram-based Amharic data ingestion, annotation, and model training pipeline for e-commerce and micro-lending insights. This project supports Named Entity Recognition (NER) in Amharic and vendor analytics using scraped messages from Ethiopian-based Telegram vendors. --- ## 📚 Table of Contents - 🛍️ Amharic E-commerce Data Extractor - 📚 Table of Contents - 📌 Project Overview - 💻 Technology Stack - ⚙️ System Requirements - 📁 Project Structure - 🧠 Task Summaries - ✅ Task 1: Data Collection \& Preprocessing - ✅ Task 2: Labeling in CoNLL Format - ✅ Task 3: Fine-Tune NER Model - ✅ Task 4: Model Comparison - ✅ Task 5: Model Interpretability - ✅ Task 6: Vendor Scorecard for Lending - 🔖 Annotation Schema - 📊 Demo Notebooks - 🚀 Getting Started - 📌 License - This project is intended for academic and research use. - 🙌 Acknowledgements --- ## 📌 Project Overview This project was built to: - Ingest and preprocess Amharic-language Telegram e-commerce data - Annotate and fine-tune models for Named Entity Recognition (NER) - Evaluate and compare multilingual models for Amharic NER - Interpret and explain NER model predictions - Build a FinTech-ready vendor analytics and lending scorecard --- ## 💻 Technology Stack - **Python 3.10+** - **Telethon** – Telegram data scraping - **HuggingFace Transformers** – Fine-tuning multilingual NER models - **spaCy, Pandas, Regex** – Preprocessing and evaluation - **SHAP & LIME** – Model interpretability - **Google Colab / Jupyter Notebooks** – Development environment - **GitHub Actions** – CI/CD and version control --- ## ⚙️ System Requirements - Python 3.10+ - pip / conda - Telegram API credentials - Access to GPU (Colab Pro, local CUDA, or AWS) --- ## 📁 Project Structure ``` ├── .github/workflows/ # CI/CD workflows ├── .vscode/ # VSCode workspace config ├── app/ # Web app backend (future use) ├── config/ │ └── config.yaml …