Sentiment_Analysis_System

Real-time sentiment prediction using Fine-Tuned BERT + Streamlit + MongoDB Atlas

---

🚀 Live Demo

🔗 App Link: https://review-sentiment-analysis-system.streamlit.app/
🔗 Model Link: https://huggingface.co/Mathusan0912/sentiment-final-model

---

📌 Project Overview

This project is a full-stack AI-powered sentiment analysis system designed to analyze customer reviews in real time.
It helps companies understand whether customers feel Positive, Neutral, or Negative about their services.

Users submit reviews → AI model predicts sentiment instantly → Review is stored in MongoDB →
Admin dashboard shows overall sentiment and latest processed reviews.

This project demonstrates real-world machine learning deployment, combining:

• Fine-tuned BERT (DistilBERT)
• Streamlit web app
• MongoDB Atlas cloud database
• HuggingFace Hub model hosting
• Streamlit Cloud deployment

---

🎯 Key Features

📝 1. User Review Submission

• Clean UI for users to submit any review.
• AI model predicts:
  • 👍 Positive
  • 😐 Neutral
  • 👎 Negative
• Shows prediction instantly with a simple thank-you message.
• Review is stored in MongoDB for backend analysis.

---

📊 2. Admin Dashboard

The admin panel provides real-time insights:

✔ Overall Sentiment Summary

Shows:

• Total reviews processed
• Positive count
• Neutral count
• Negative count

Useful for understanding customer satisfaction quickly.

✔ Latest 20 Processed Reviews

• Displays a live table with the 20 most recent reviews.
• Includes predicted sentiment for each.
• Helps companies understand trends & customer issues immediately.

✔ Automatic Model Processing

• Backend automatically processes all unprocessed reviews.
• Ensures no review is missed.

---

🧠 AI Model Details

• Base model: distilbert-base-uncased
• Fine-tuned on 30,000 balanced sentiment samples
• Label mapping:
  • 0 → Negative
  • 1 → Neutral
  • 2 → Positive
• Model is hosted on HuggingFace Hub
• Loaded by Streamlit using:

AutoTokenizer.from_pretrained("Mathusan0912/sentiment-final-model")
AutoModelForSequenceClassification.from_pretrained("Mathusan0912/sentiment-final-model")