Sentiment Analysis using LSTM and GRU

This project focuses on classifying sentiments (positive/negative) from textual data using Recurrent Neural Networks (RNN) models, specifically LSTM and GRU, implemented in TensorFlow and Scikit-learn. The project leverages the IMDB dataset to train and evaluate the models.

Table of Contents

• Project Overview
• Technologies Used
• Dataset
• Model Performance
• Installation
• Usage
• Results
• Future Work
• Contributing

Project Overview

This project demonstrates how to use LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit) networks for sentiment classification. Key steps include:

• Preprocessing text data: Sequence padding, tokenization, and embedding layers are employed.
• Model training: Both LSTM and GRU models are trained on the IMDB dataset.
• Evaluation: Model accuracy and loss are analyzed to gauge performance.

Key Achievements

• Boosted sentiment classification accuracy to 87.15% using LSTM, reducing loss to 0.3507.
• Achieved 87.77% accuracy by employing GRU, optimizing model performance.
• Improved model efficiency and training time by 10% through sequence padding, dropout regularization, and embeddings.

Technologies Used

• Python: Core language for data manipulation and model building.
• TensorFlow: For building and training the LSTM and GRU models.
• Scikit-learn: For additional machine learning utilities.
• Pandas: For handling and preprocessing the dataset.
• Matplotlib/Seaborn: For visualizing training performance.

Dataset

The IMDB dataset is a widely used dataset for binary sentiment classification (positive or negative). It contains 25,000 highly polar movie reviews for training and 25,000 for testing.

The dataset can be easily loaded using:

from tensorflow.keras.datasets import imdb

Model Performance

1. LSTM: Achieved 87.15% accuracy with a loss of 0.3507.
2. GRU: Achieved 87.77% accuracy with further optimizations.

The models were trained using sequence padding, embedding layers, and dropout regularization to avoid overfitting.

Installation

Clone the repository and install the necessary dependencies:

git clone https://github.com/your-username/Sentiment-Analysis.git
cd Sentiment-Analysis
pip install -r requirements.txt

The requirements.txt file contains all the necessary Python libraries for this project.

Usage

To run the training script for the LSTM model:

python train_lstm.py

For the GRU model:

python train_gru.py

Results

LSTM Results:

• Accuracy: 87.15%
• Loss: 0.3507

GRU Results:

• Accuracy: 87.77%
• Loss: Improved, with optimized training time and performance.

Training loss and accuracy over epochs can be visualized using the included scripts. Example:

import matplotlib.pyplot as plt

# Plot training & validation accuracy
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train', 'Validation'], loc='upper left')
plt.show()

Future Work

• Explore attention mechanisms to further improve accuracy.
• Experiment with BERT or transformer models for even better performance.
• Deploy the model using Flask or FastAPI as a sentiment analysis web service.

Contributing

Feel free to fork this project, make changes, and submit a pull request. Contributions are welcome!