joss413/amharic-ner

Amharic Named Entity Recognition using AfroXLM-R # Amharic Named Entity Recognition An AI pipeline that classifies customer complaints, extracts structured ticket data, generates empathetic responses, and logs everything to a searchable dashboard. --- ## 🚀 Live Demo 👉 Try it on Hugging Face Spaces --- ## What It Does 1. Accepts any Amharic sentence as input 2. Tokenizes text using AfroXLM-R subword tokenizer 3. Predicts named entity tags for each token 4. Highlights entities with color-coded chips: 🔴 PER — Person names (ሰው) 🟢 LOC — Locations (ቦታ) 🔵 ORG — Organizations (ድርጅት) 🟡 DATE — Dates (ቀን) 5. Displays entity summary cards with counts --- ## Screenshots | | | |:---:|:---:| | | | | *App Interface* | *Person & Location Detection* | | | | | *Organization & Date Detection* | *Entity Summary Cards* | --- ## Tech Stack - Python 3.10 - PyTorch — deep learning framework - HuggingFace Transformers — AfroXLM-R model - AfroXLM-R-base (Davlan/afro-xlmr-base) — pretrained on 17 African languages including Amharic - MasakhaNER 1.0 — Amharic NER dataset - Gradio — web interface - seqeval — NER evaluation metrics --- ## Architecture ``` amharic-ner/ │ ├── app.py # Gradio web interface ├── requirements.txt # Dependencies ├── images/ # Screenshots │ ├── screenshot1.png │ ├── screenshot2.png │ ├── screenshot3.png │ └── screenshot4.png └── saved_model/ # Trained model (hosted on HuggingFace) ├── config.json ├── pytorch_model.bin ├── tokenizer files └── ner_config.json ``` --- ## Evaluation ``` (AfroXLM-R-base): 0.76 ``` --- ## Running Locally ```bash # Clone the repository git clone github.com cd amharic-ner # Install dependencies pip install -r requirements.txt # Run the app python app.py ```