Beya2123/Amharic-News-Classification-with-Multilingual-Transformers.

Amharic News Classification with Multilingual Transformers. This repository contains three transformer-based models for classifying Amharic news articles. The models leverage state-of-the-art multilingual language models to effectively handle the Amharic language. # Amharic-News-Classification-with-Multilingual-Transformers. Amharic News Classification with Multilingual Transformers. This repository contains three transformer-based models for classifying Amharic news articles. The models leverage state-of-the-art multilingual language models to effectively handle the Amharic language. Amharic News Classification with Multilingual Transformers This repository contains three transformer-based models for classifying Amharic news articles. The models leverage state-of-the-art multilingual language models to effectively handle the Amharic language. 📋 Project Overview This project implements and compares three different multilingual transformer models for Amharic news classification: mBERT (Multilingual BERT) XLM-RoBERTa (Cross-lingual Language Model) Afro-XLM-RoBERTa (African-focused multilingual model) These models are trained to classify Amharic news articles into various categories (politics, sports, technology, etc.) using the Amharic News Dataset. 🚀 Models 1. mBERT (Multilingual BERT) Model: bert-base-multilingual-cased Description: Google's BERT model trained on 104 languages including Amharic File: mBert.py Key Features: 12-layer, 768-hidden, 12-heads, 110M parameters Trained on Wikipedia data Supports tokenization for Amharic 2. XLM-RoBERTa Model: xlm-roberta-base Description: Facebook's cross-lingual model trained on 100 languages File: XLM-roberta.py Key Features: 12-layer, 768-hidden, 12-heads, 270M parameters Trained on CommonCrawl data Specifically designed for cross-lingual tasks 3. Afro-XLM-Roberta Model: castorini/afroxlmr-base Description: XLM-RoBERTa model fine-tuned on African languages File: Afro-xlmr-base.py Key Features: Optimized for African languages including Amharic Better performance on low-resource languages Includes language-specific adaptations 📊 Dataset The models are trained on the Amharic News Dataset which contains: Amharic news articles with category labels Multiple c …