Fake Review Detection Chrome Extension

A lightweight Chrome extension that identifies fake reviews on e-commerce platforms like Amazon and Flipkart using a Bi-LSTM model served via a scalable cloud API.

Overview

This extension enhances online shopping by detecting potentially fake reviews using machine learning and NLP. It scrapes visible reviews from product pages, sends them to a Flask-based backend hosted on AWS ECS, and classifies them using a trained Bi-LSTM model.

Features

• Real-time review classification (Fake / Real)
• Bi-LSTM-based NLP model trained on review text + metadata
• Cloud-hosted backend (Flask + Docker + AWS ECS)
• DOM scraping for Amazon and Flipkart pages
• Fast, asynchronous API calls for real-time inference
• Minimal, responsive UI popup (React + TypeScript)

---

Tech Stack

Layer	Tech
Frontend	TypeScript, Chrome Extension APIs
ML Backend	Python, Flask, TensorFlow, Bi-LSTM, Scikit-learn
Infrastructure	Docker, AWS ECS (Fargate)
Communication	REST API (JSON over HTTP)

---

Model Details

• Architecture: Bi-directional LSTM
• Input: Tokenized review text + reviewer metadata
• Accuracy: 97% on test data
• Deployment: TensorFlow Serving behind a Flask API

---

How It Works

1. User browses a product page
2. Extension scrapes visible reviews from the DOM
3. API sends text → ML backend → prediction
4. Results are displayed inline or in the popup

---

graph TD
    subgraph User Browser
        A1[User visits product page ,Amazon / Flipkart]
        A2[Chrome Extension]
        A3[Content Script scrapes reviews]
        A1 --> A2 --> A3
    end

    A3 --> B1[REST API Request, Review Data]

    subgraph Backend [Cloud Backend AWS ECS]
        B1 --> B2[Flask API Server]
        B2 --> B3[Bi-LSTM ML Model]
        B3 --> B4[Prediction: Real or Fake]
    end

    B4 --> C1[API Response with Classification]

    subgraph Extension UI
        C1 --> D1[Popup UI Displays Classification]
    end

Loading

Future Improvements

• Add model confidence visualization
  Display prediction confidence alongside classification, e.g., "78% Fake" or "91% Real".

• Improve domain support
  Extend support beyond Amazon and Flipkart to other marketplaces like Snapdeal, Meesho, or international sites like eBay.

• Track user feedback for active retraining
  Allow users to confirm or correct predictions, and feed that data into a retraining pipeline to continuously improve the model.

License

This project is licensed under the MIT License.