Amharic Hate Speech Detection

Deep learning NLP system using fine-tuned multilingual BERT (mBERT) to classify and detect hate speech in Amharic.

Amharic Hate Speech Detection is a specialized Natural Language Processing (NLP) system developed as Natnael’s B.Sc. Electrical & Computer Engineering final-year project at Addis Ababa University. It detects and mitigates online hate speech and abusive language in the Amharic language using transfer learning on fine-tuned multilingual BERT (mBERT).

Key Capabilities

  • mBERT Fine-Tuning: Built on top of pre-trained multilingual BERT representations, adapted specifically for Amharic morphological and syntactical patterns.
  • Hugging Face Model Hub: Pre-trained and fine-tuned weights published and hosted on Hugging Face Hub (NathyB/Hate-Speech-Detection-in-Amharic-Language-mBERT).
  • Content Moderation Pipeline: Provides reproducible benchmarking for hate speech, offensive discourse, and neutral sentiment classification.
  • Tech Stack: Python, PyTorch, Hugging Face Transformers, Datasets, mBERT, Jupyter/Colab.
Hugging Face Model Jupyter Notebook View GitHub Repository