Emnet-tes/Amharic-Ecom-NER

End-to-end NER pipeline for Amharic e-commerce vendor analysis. Extracts entities from Telegram data to generate risk insights for micro-lending in Ethiopia’s FinTech sector. # 🛒 Amharic E-commerce NER & FinTech Analytics **A comprehensive Named Entity Recognition (NER) pipeline for Amharic e-commerce data with practical FinTech applications for micro-lending decisions.** ## 🎯 Project Overview This project implements an end-to-end NER pipeline specifically designed for Amharic e-commerce text analysis, culminating in a vendor analytics system for micro-lending assessment. The system processes Telegram channel data from Ethiopian e-commerce vendors to extract business entities and calculate lending risk scores. ## ✅ Completed Tasks ### **Task 1: Project Structure & Data Preprocessing** ✅ - **Objective**: Establish organized project structure and preprocess Telegram data - **Deliverables**: - Clean project directory with notebooks, data, and models folders - Preprocessed CSV and JSON datasets with cleaned and tokenized messages - **Status**: ✅ Complete ### **Task 2: CoNLL Data Labeling** ✅ - **Objective**: Convert raw Amharic text to CoNLL format with entity annotations - **Deliverables**: - `conll_data_labeling.ipynb` - Automated labeling system - CoNLL formatted training data with B-I-O tagging scheme - Entity types: PRODUCT, PRICE, LOCATION - **Status**: ✅ Complete ### **Task 3: Model Fine-tuning** ✅ - **Objective**: Fine-tune transformer models for Amharic NER - **Deliverables**: - `model_fine_tuning.ipynb` - Complete training pipeline - Fine-tuned AfroXLMR model achieving F1: 0.3939 - Model comparison across multiple architectures - **Status**: ✅ Complete ### **Task 4: Model Comparison** ✅ - **Objective**: Evaluate and compare different NER models - **Deliverables**: - `model_comparison.ipynb` - Comprehensive evaluation framework - Performance metrics for DistilBERT, XLM-RoBERTa, and AfroXLMR - Best model selection (AfroXLMR) based on F1 scores - **Status**: ✅ Complete ### **Task 5: Model Interpretability** ✅ - **Objective**: Analyze model predictions and provide interpretability insights - **Deliverables**: - `model_int …