meron-23/Ethiopian-Ecommerce-Entity-Extractor

# Ethiopian-Ecommerce-Entity-Extractor A machine learning pipeline for extracting product names, prices, and locations from Amharic-language Telegram e-commerce channels using Named Entity Recognition (NER). ## Features - Telegram message scraping - Amharic text cleaning and tokenization - CoNLL-style NER labeling - Fine-tuning multilingual LLMs (XLM-R, mBERT) - OCR for text embedded in images - SHAP and LIME for model explainability ## Directory Structure - `data/`: Raw and processed text/images - `src/`: All training, preprocessing, scraping logic - `notebooks/`: Training and analysis experiments ## Setup ```bash git clone github.com cd Ethiopian-Ecommerce-Entity-Extractor pip install -r requirements.txt