abush26/Sentiment-Analysis-NLP

A Python tool for Amharic text sentiment analysis with fine-tuned and translation-based approaches, featuring a web interface and programmable API. # Amharic Sentiment Analysis ## Overview This repository contains a Python-based tool for performing sentiment analysis on Amharic text. Amharic is the official language of Ethiopia and one of the Semitic languages spoken in the Horn of Africa region. ## Features * **Sentiment classification**: Classify Amharic text as positive, negative, neutral, very positive, or very negative * **Polarity scoring**: Show confidence scores for each sentiment category * **Two implementation approaches**: * Fine-tuned model specifically trained on Amharic data * Translation + zero-shot classification pipeline for quick implementation ## Demo The project includes a Gradio-based web interface that demonstrates the functionality: 1. Enter Amharic text in the input field 2. Click "Analyze Sentiment" 3. View the translated English text and detailed sentiment analysis results ## Implementation Approaches ### 1. Fine-Tuned Model Approach The primary approach uses a model fine-tuned specifically on Amharic sentiment data: * **Word Vectors**: Uses FastText embeddings which have shown better results for the Amharic language * **Training Data**: Custom dataset of labeled Amharic text (available in the `data` folder) * **Model Architecture**: Neural network with embedding layer using FastText vectors Note: The FastText model is not included in this repository due to its large size but can be found on the FastText website. ### 2. Quick Implementation Approach For users who prefer a simpler implementation without fine-tuning details, the `hugging_face` folder provides a translation-based pipeline: 1. **Translation Step**: Convert Amharic text to English using Facebook's NLLB (No Language Left Behind) model 2. **Classification Step**: Apply zero-shot classification on the translated English text using DeBERTa-v3-base-mnli-fever-anli This approach requires less setup and domain expertise while still providing reasonable results. ## How …