Fake-Review-Detection

📌 Overview

This repository contains an LSTM-based deep learning model for Fake-Review-Detection using pre-trained GloVe word embeddings. The project involves:

• Preprocessing: Loading GloVe embeddings and preparing an embedding matrix.
• Model Training: Implementing an LSTM model with hyperparameter tuning using Ray Tune.
• Hyperparameter Optimization: Tuning LSTM layers, units, and learning rate with Bayesian Optimization.
• Evaluation: Assessing model performance with accuracy, binary cross-entropy loss, and confusion matrices.

🚀 Live Demo

Try the deployed model here: Fake Review Detection App

📂 Project Structure

📦 LSTM-Text-Classification
├── 📜 Embeddings.ipynb        # Loading GloVe embeddings and preparing embedding matrix
├── 📜 Hyperparameter Tuning.ipynb # Hyperparameter tuning using Ray Tune
├── 📜 Model Training.ipynb    # Training and evaluating the model
├── 📜 glove.6B.50d.txt        # Pre-trained GloVe embeddings (50d vectors)
├── 📜 README.md               # Project documentation
└── 📜 requirements.txt        # Dependencies and packages required

🔬 Model Architecture

• Embedding Layer: Pre-trained GloVe embeddings (non-trainable)
• LSTM Layers: Multiple LSTM layers with tuned units
• Dense Layer: Fully connected layer with a sigmoid activation function
• Optimizer: Adam with a tuned learning rate
• Loss Function: Binary Cross-Entropy

📊 Evaluation

The model is evaluated using:

• Accuracy
• Binary Cross-Entropy Loss
• Confusion Matrix Visualization

A sample confusion matrix:

                Predicted
            Machine  |  Human
True  Machine    [TN]  |   [FP]
      Human     [FN]  |   [TP]

🛠️ Hyperparameter Tuning

Hyperparameter optimization is performed using Ray Tune with Bayesian Optimization (BOHB). The tuning process explores:

• Number of LSTM layers
• LSTM hidden units
• Learning rate

📌 Dependencies

• Python 3.8+
• TensorFlow
• NumPy
• Pandas
• Ray Tune
• Matplotlib
• Seaborn