Sentiment Analysis API

Overview

This project builds a sentiment classifier on Amazon product reviews and serves it as a FastAPI Rest API.

Tech Stack

• Python
• Scikit-learn (Naive Bayes Classifier)
• FastAPI( Rest API)
• Docker (containerization)

Features

• Post/Predict -Accepts review text and return sentiment (Positive/negative/neutral)

Project Structure

• notebooks/ :Model training code
• app/ : FastAPI app saved model
• Dockerfile/ : Container for deployment

Example request

curl -x Post http://localhost:8000/predict/ -H "Content-Type: application/json" -d '{"text": This product is amazing!}'

Results

• Macro ROC-AUC: 0.86
• Model: TF-IDF + Logistic Regression (class_weight=balanced)
• confusion matrix:

Challenges

• Dataset highly imbalanced ( Positive reviews dominated)
• Addressed via class weights

Future Improvements

• Apply oversampling(SMOTE)
• Tray transformer model (DistilBERT/BERT)