rufaeleshetu/amharic-ecommerce-ner-ethiomart

Fine-tuned Amharic NER model for extracting vendor data from Telegram e-commerce channels to support FinTech micro-lending decisions. # Amharic E-commerce NER for EthioMart This project fine-tunes Amharic NER models on Telegram-based e-commerce data to extract products, prices, and locations. The goal is to help EthioMart identify top vendors for micro-lending. ## Tasks - Data scraping from Telegram - Amharic NER labeling (CoNLL format) - Model fine-tuning (XLM-Roberta, AfroXLMR, etc.) - Model interpretability (SHAP, LIME) - Vendor scoring system for FinTech lending ## Folder Structure - `data/` - raw and cleaned Telegram data - `notebooks/` - notebooks for scraping, training, and evaluation - `models/` - saved fine-tuned models - `utils/` - Python helpers (e.g., tokenizer, cleaning) - `vendor_scorecard/` - scripts to score vendors