aymenhayderhtml-pixel/smart-amharic-predictive-keyboard

Amharic predictive text keyboard with next-word suggestion using N-gram + Bi-LSTM models. Built for faster typing in አማርኛ. # ⌨️ Smart Amharic Predictive Keyboard **Amharic next-word prediction keyboard** built with a **Trigram N-gram Model**. A practical Machine Learning mini-project that helps users type faster in አማርኛ by suggesting the next word based on typing history. ## ✨ Features - Real-time next-word suggestions (top 5) - Clickable suggestion buttons (just like phone keyboards) - Trained on **765,167** real Amharic news sentences - Large vocabulary (~555k words) - Clean and responsive Streamlit web interface - Supports full Ge'ez (Amharic) script ## 🛠️ Tech Stack - **Python** - **N-gram (Trigram)** language model with backoff - **Pandas** + **Regex** for Amharic text preprocessing - **Streamlit** for the interactive keyboard demo ## 📁 Project Structure ``` smart-amharic-predictive-keyboard/ ├── data/ │ ├── raw/ # (ignored - contains Amharic News Dataset.csv) │ └── processed/ │ └── cleaned_sentences.txt ├── src/ │ ├── preprocess.py │ └── ngram_model.py ├── app/ │ └── main.py # Streamlit keyboard demo ├── models/ │ └── amharic_trigram.pkl # Trained model (run training script to generate) ├── notebooks/ ├── requirements.txt └── README.md ``` ## 🚀 How to Run ### 1. Clone the repo ```bash git clone github.com cd smart-amharic-predictive-keyboard ``` ### 2. Install dependencies ```bash pip install -r requirements.txt ``` ### 3. Add the dataset Download the Amharic News Text Classification Dataset and place the CSV in: ``` data/raw/Amharic News Dataset.csv ``` ### 4. Preprocess the data ```bash python src/preprocess.py ``` ### 5. Train the model (only once) ```bash python src/ngram_model.py ``` ### 6. Run the keyboard ```bash streamlit run app/main.py ``` --- ## 📊 Dataset - **Source**: Amharic News Text Classification Dataset - **Size**: 51,483 news articles → 765,167 cleaned sentences - **Used for**: Training the trigram language model --- # …