Trust Score – AI-Powered Shopping Assistant

The Trust Score Extension helps you shop smarter on Amazon by automatically analyzing product listings with AI.
It detects signs of counterfeit products, fake reviews, misleading descriptions, and suspicious return patterns — then gives you a simple Trust Score (%) directly on the product page.

---

Features

Feature	Description
Trust Score Badge	Instant percentage score on every product
Image-Text Consistency	Uses CLIP + BLIP to validate alignment between product images and descriptions
Review Authenticity	Uses BART-MNLI to detect templated, bot-like, or suspicious reviews
Brand Verification	Logo and branding cross-checks (heuristics + vision models)
Return Feedback Analysis	Uses DistilBERT to classify customer complaints (e.g., fake, wrong item, poor quality)
Lightweight & Fast	Works seamlessly while browsing
Privacy Friendly	Sends only essential product data to the AI API

---

Architecture

---

Machine Learning Models Used

Task	Model	Purpose
Image–Text Similarity	openai/clip-vit-base-patch32	Embeds product images & text, measures alignment
Image Captioning	Salesforce/blip-image-captioning-base	Generates captions from product images to validate against product title/description
Review Authenticity	facebook/bart-large-mnli	Zero-shot classification of reviews to detect authenticity & coherence
Returns Feedback	distilbert/distilbert-base-uncased	Sentiment & intent analysis on return complaints

---

Installation

1. Download the Extension

1. Go to the Releases section of this repo.
2. Download the latest trust-score-extension.zip.
3. Extract the zip file to a folder on your computer.

2. Load in Chrome

1. Open Chrome (or any Chromium-based browser).
2. Go to chrome://extensions/.
3. Enable Developer Mode (top-right).
4. Click Load unpacked and select the "extension" folder.
   Done — the extension runs automatically on Amazon product pages!

---

How to Use

1. Visit any Amazon product page.
2. Wait a moment while the extension analyzes the product.
3. A Trust Score button will appear at the bottom-left.
4. Hover or click to see a detailed modal with analysis.

Example Trust Scores

Score	Meaning
95%	Reliable listing
60%	Some risks detected
30%	High chance of counterfeit or misleading info

---

Technical Architecture

1. Chrome Extension (Frontend)

   • Extracts product info (title, description, images, reviews).
   • Sends JSON payload to backend (/trust_score).
   • Renders Trust Score badge + modal UI via Shadow DOM.

2. Backend (FastAPI-based Orchestrator)

   • Receives product data.
   • Runs multiple ML modules: CLIP + BLIP for images, BART-MNLI for reviews, DistilBERT for returns.
   • Aggregates module scores into a final Trust Score.
   • Returns JSON with per-module details & explanations.

3. ML Models (Hugging Face)

   • CLIP: image-text similarity
   • BLIP: image captioning
   • BART-MNLI: zero-shot review authenticity
   • DistilBERT: return sentiment classification

4. Response Returned to Extension

   {
     "trust_score": 0.82,
     "details": {
       "image_text_alignment": {"score":0.9,"model":"clip-vit-base-patch32","summary":"Images match description"},
       "review_authenticity": {"score":0.7,"model":"bart-large-mnli","summary":"Some templated reviews detected"},
       "logo_verification": {"score":0.95,"summary":"Brand logo verified"},
       "returns_feedback": {"score":0.5,"model":"distilbert-base-uncased","summary":"Multiple returns cite authenticity issues"}
     }
   }