farahhussei10-arch/AI-Multilingual-Sentiment-Analysis

AI for language analysis for three languages: Somali, English, Kiswahili # English, Kiswahili & Somali AI Sentiment Analysis Tool A professional portfolio project that classifies user text in English, Kiswahili, and Somali as **positive**, **negative**, or **neutral** sentiment. ## Overview This project demonstrates a complete NLP workflow with a reusable model and REST API: - dataset preparation - text preprocessing - TF-IDF vectorization - logistic regression training - model persistence - API inference with FastAPI - automated tests for reliability ## Problem Statement Organizations need a reliable way to classify short customer or service text in English, Kiswahili, and Somali. This tool provides a lightweight sentiment classifier that can be run locally and explained in an interview. ## Objectives - Build a genuinely working sentiment analysis application - Support English, Kiswahili, and Somali text - Use explainable, reproducible ML components - Provide a REST API for predictions - Include tests and documentation for GitHub readiness ## Technologies Used - Python 3 - pandas - NumPy - scikit-learn - FastAPI - Uvicorn - pytest - HTML, CSS, JavaScript ## Project Architecture ``` sentiment-analysis/ │ ├── data/ │ └── sentiment_dataset.csv ├── models/ │ ├── sentiment_model.pkl │ └── tfidf_vectorizer.pkl ├── src/ │ ├── preprocessing.py │ ├── train.py │ ├── predict.py │ └── api.py ├── tests/ │ ├── test_api.py │ ├── test_predict.py │ └── test_preprocessing.py ├── app.py ├── requirements.txt ├── README.md └── .gitignore ``` ## Dataset Description The dataset is a small, custom training dataset stored in `data/sentiment_dataset.csv`. It contains: - `text`: English, Kiswahili, or Somali sentences - `language`: language code (`en`, `sw`, or `so`) - `sentiment`: sentiment label (`positive`, `negative`, `neutral`) The dataset currently includes 30 English examples, 30 Kiswahili examples, and 9 Somali examples. It is intentionally compact and created for demonstrative portfolio purposes. It is not a claim of lar …