# EthioMart Lite: Amharic NER for Telegram E-Commerce
📌 **Project Overview**
**EthioMart Lite** is a lightweight pipeline that transforms Telegram e-commerce messages in Amharic into structured data for Named Entity Recognition (NER). The solution includes:
* 🧲 Telegram message scraping
* 🧹 Amharic-English text preprocessing
* 🏷️ Manual token-level labeling in CoNLL format
* ⚙️ NER model training and interpretability
* 📈 Vendor analytics for business intelligence
---
## 🏆 Key Achievements
### ✅ Data Pipeline
* Scraped **1,000+ messages** from 5 Amharic Telegram vendors:
* `ZemenExpress`, `nevacomputer`, `helloomarketethiopia`, `Fashiontera`, `kuruwear`
* Saved in `raw_telegram_data.json`
### ✅ Data Processing
* Text cleaned, tokenized, and exported to `preprocessed_data.csv`
* Nulls, emojis, and links removed
* Script: `preprocess_data.ipynb`
### ✅ Manual NER Labeling
* 30 messages (\~400+ tokens) labeled using BIO format
* Output in `labeled_data.conll`
* Script: `label_data_to_conll.ipynb`
### ✅ NER Modeling
* Transformer models fine-tuned for NER
* Performance benchmarked in `model_comparision.ipynb`
* Interpretability analysis in `model_interpretability.ipynb`
### ✅ Vendor Analytics
* Created vendor scoring logic based on:
* Views per post
* Posting frequency
* Price profile
* Script: `vendor_scorecard.ipynb`
---
## 📂 Repository Structure
```plaintext
Amharic-E-commerce-Data-Extractor/
├── .github/workflows/ # GitHub Actions
├── notebooks/
│ ├── scrape_telegram.ipynb # Task 1 - Scraping
│ ├── preprocess_data.ipynb # Task 2 - Preprocessing
│ ├── label_data_to_conll.ipynb # Task 3 - Manual labeling
│ ├── model_training.ipynb # Task 4 - Fine-tuning models
│ ├── model_comparision.ipynb # Task 5 - Benchmarking
│ ├── model_interpretability.ipynb # Task 6 - SHAP/LIME insights
│ └── vendor_scorecard.ipynb # Task 6 - Vendor profiling
│
├── requirements.txt # Pytho …