Nehmyabiruk/EthioMart-Amharic-NER

# EthioMart Amharic NER **Named Entity Recognition for Amharic E-commerce Telegram Messages** Fine-tuned **XLM-RoBERTa** model to extract key entities (**Product**, **PRICE**, **LOC**) from Amharic Telegram e-commerce posts — powering vendor analytics and micro-lending decisions for **EthioMart**. ## 🎯 Project Goal Transform messy Telegram vendor messages into structured data to: - Centralize e-commerce information from multiple channels - Identify promising vendors for micro-lending - Support EthioMart's vision of becoming Ethiopia's unified Telegram e-commerce hub **Target entities**: - **Product** (B-Product / I-Product) - **PRICE** (B-PRICE / I-PRICE) - **LOCATION** (B-LOC / I-LOC) ## ✨ Features - Telegram message scraping from 5+ Ethiopian e-commerce channels - Text cleaning & basic rule-based pre-annotation - Manual labeling in CoNLL format - Fine-tuning XLM-RoBERTa for Amharic NER - Per-class & overall evaluation metrics (seqeval) - Automatic performance bar chart popup after training ## 📂 Project Structure ethiomart-amharic-ner/ ├── src/ │ ├── scraper.py # Collects messages from Telegram channels │ ├── preprocessing.py # Cleans text & generates pre-annotated CSV │ └── train_Ner.py # Full NER training & evaluation pipeline ├── data/ │ ├── raw/ │ │ └── telegram_raw.csv # Raw scraped messages (channel, text, views, date...) │ ├── processed/ │ │ └── annotated_ner.csv # Pre-annotated or manually corrected data │ └── labels/ │ ├── train.conll # Training set in CoNLL format │ └── valid.conll # Validation set ├── models/ │ └── xlm-roberta-amharic-ner-final/ # Saved fine-tuned model (not in git) ├── requirements.txt ├── .gitignore └── README.md ## 🚀 Quick Start ### 1. Clone the repository ```bash git clone github.com cd EthioMart-Amharic-NER 2. Install dependencies Bashpip install -r requirements.txt requirements.txt c …