A browser-based NLP application that analyzes product reviews using an interpretable rule-based scoring engine to identify potentially fake, spam, or incentivized reviews. The application runs entirely in the browser and provides transparent explanations for every prediction.
Home Page
|
Genuine Review Detection
|
Fake Review Detection
|
Online reviews play a major role in influencing consumer decisions, making fake and incentivized reviews an increasing challenge for e-commerce platforms.
ReviewGuard demonstrates how Natural Language Processing (NLP) concepts can be applied through a transparent, rule-based scoring system. Instead of relying on a trained machine learning model, the application evaluates multiple linguistic patterns and combines them into a Fake Probability Score (0–100).
Each prediction is accompanied by a detailed explanation, allowing users to understand exactly why a review has been classified as Genuine, Suspicious, or Fake.
Note
ReviewGuard is an educational project demonstrating rule-based NLP techniques for fake review detection. It is designed to illustrate interpretable text analysis and is not intended to replace production-grade machine learning systems.
- Rule-based NLP scoring engine
- Multi-signal linguistic analysis
- Fake Probability Score (0–100)
- Interactive confidence meter
- Explanation of detected linguistic signals
- Session history with one-click review reload
- Live review statistics
- Fully responsive interface
- Runs entirely in the browser
- No backend or external dependencies
- User enters a product review.
- The review is preprocessed.
- Eight linguistic signals are analyzed.
- A weighted score is calculated.
- A Fake Probability Score is generated.
- The review is classified and the detected signals are explained.
| Signal | Description |
|---|---|
| Excessive punctuation | Detects repeated exclamation marks |
| ALL CAPS words | Detects exaggerated emphasis |
| Superlative phrases | Identifies unrealistic promotional language |
| Generic recommendation phrases | Detects common spam patterns |
| Repeated positive adjectives | Detects unnatural repetition |
| Very short reviews | Flags low-information reviews |
| Personal experience indicators | Reduces the fake score for authentic experiences |
| Balanced language | Rewards constructive criticism and balanced opinions |
| Fake Probability | Verdict |
|---|---|
| 0–29 | ✅ Genuine |
| 30–54 | |
| 55–100 | 🚩 Fake |
User Review
│
▼
Text Preprocessing
│
▼
Rule-Based NLP Engine
│
▼
Weighted Score Calculation
│
▼
Fake Probability Score
│
▼
Verdict Generation
│
▼
Interactive User Interface
Frontend
- HTML5
- CSS3
- JavaScript (ES6)
Deployment
- GitHub Pages
Fonts
- Google Fonts
ReviewGuard/
│
├── images/
│ ├── home.png
│ ├── genuine.png
│ └── fake.png
│
├── index.html
├── report.docx
└── README.md
OMG!!! BEST PRODUCT EVER!!! Absolutely AMAZING!!! Everyone should buy this right now!! HIGHLY RECOMMEND!!!
I bought this last month for my daughter. The build quality is good, although the battery life could be better. Delivery took five days.
- TensorFlow.js integration
- Transformer-based NLP models (BERT / DistilBERT)
- Browser extension for real-time review analysis
- Batch review processing
- Multilingual support
- Explainable AI visualizations
A detailed project report is included in this repository covering:
- Problem Statement
- Literature Review
- System Design
- Implementation
- Testing
- Results
- Future Scope
Debaditya Ghosal
Computer Science Undergraduate (Artificial Intelligence & Machine Learning)
This project is intended for academic and educational purposes.