ReviewLens 🕵️‍♂️

A hybrid AI-powered web application designed to detect deceptive hotel reviews using a combination of Semantic NLP (TF-IDF) and Stylometric Analysis.

🚀 Overview

Fake reviews are a significant problem in the digital economy. This project uses machine learning to distinguish between truthful and deceptive reviews by analyzing not just what is said, but how it is written.

Key Features:

• Hybrid Detection Engine: Combines TF-IDF vectorization with handcrafted stylometric features.
• Stylometric Metrics:
  • Vocabulary Diversity: Measures the richness of the reviewer's vocabulary.
  • Punctuation Intensity: Tracks excessive use of exclamation and question marks.
  • Personal Pronoun Ratio: Analyzes the self-focus of the reviewer (often higher in fake reviews).
• Interactive Web Interface: Built with Flask for real-time analysis.

---

🛠️ Tech Stack

• Backend: Python, Flask
• Machine Learning: Scikit-learn, Joblib
• NLP: NLTK, Regex
• Frontend: HTML5, CSS3 (Vanilla)
• Data Handling: Pandas, NumPy, Scipy

---

📂 Project Structure

├── data/               # Datasets used for training
├── models/             # Saved ML models (.pkl)
├── notebooks/          # Jupyter notebooks for EDA and Training
├── templates/          # HTML templates for the Flask app
├── app.py              # Main Flask application
├── requirements.txt    # Project dependencies
└── .gitignore          # Files excluded from version control

---

⚙️ Installation & Usage

1. Clone the repository

git clone https://github.com/Ankush-22/ReviewLens.git
cd ReviewLens

2. Create a Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Run the Application

python app.py

Open your browser and navigate to http://127.0.0.1:5000.

---

🧠 How it Works

1. Preprocessing: Text is cleaned, lemmatized, and stop-words are removed.
2. Feature Extraction:
   • TF-IDF: Captures the importance of specific words.
   • Stylometry: Calculates mathematical ratios of language use.
3. Classification: A pre-trained hybrid model (Random Forest/SVM) predicts the likelihood of the review being fake.

---