nardosdubale1064/amharic_hate_speech_detection

# 🗣️ Amharic Social Media Hate Speech Analyzer A web-based NLP tool for detecting **hate speech**, **offensive**, and **normal** content in **Amharic** text collected from public Telegram channels and groups. Built for educational and research purposes, this project demonstrates practical applications of Natural Language Processing for low-resource languages. ## 🚀 Features * **🔍 Web Interface** Simple and clean web form where users input Telegram URLs to view categorized analysis results. * **🧹 Amharic Text Preprocessing** Custom normalization for Amharic characters, removal of URLs, mentions, hashtags, emojis, and a comprehensive stopword list. * **✂️ Sentence-Level Tokenization** Breaks down long messages and comments into individual sentences for more detailed classification. * **🧠 Machine Learning Model** Logistic Regression classifier using TF-IDF features, trained on `uhhlt/amharichatespeechranlp` dataset. * **📡 Telegram Scraper (Telethon)** Asynchronously fetches: * Comments from a specific Telegram post. * Messages (and their comments) from channels or groups. * Automatically detects message type (channel/group) and supports a custom message limit (default: 1000). * **📊 Result Summary** Shows: * Number of messages/comments scraped * Total sentences analyzed * Category breakdown: hate, offensive, and normal * Example sentences for each category * **🧩 Modular Design** Code is organized for readability and reusability using separate modules. --- ## 📁 Project Structure ```plaintext amharic_hate_speech_analyzer/ ├── config.py # Configuration settings (API keys, model paths, label mapping) ├── amharic_preprocessing.py # Amharic text cleaning, normalization, and tokenization ├── model_trainer.py # Model training and saving script ├── telegram_scraper.py # Telegram data fetching logic (messages + comments) ├── server.py # Main Flask web application ├── …