Sentiment Analysis with RNN - IMDB Dataset

Project Information

• Title: Sentiment Analysis with RNN on IMDB Dataset
• Students:
  • Abhijeet (055002)
  • Jhalki Kulshrestha (055017)
• Group Number: 19

---

1. Introduction

This project focuses on building a deep learning pipeline to classify movie reviews as positive, negative, or neutral using the IMDB dataset. It leverages a Recurrent Neural Network (RNN) architecture with LSTM layers to effectively capture sentiment in sequential text data.

---

2. Dataset Description

The dataset used is the IMDB movie review dataset provided by Keras:

• Total Reviews: 50,000
• Training Set: 25,000 reviews
• Testing Set: 25,000 reviews
• Labels: Binary (1 = Positive, 0 = Negative)
• Vocabulary Size: Top 10,000 most frequent words
• Data Format: Each review is encoded as a list of word indices

---

3. Project Objectives

• Build an end-to-end NLP pipeline for sentiment classification.
• Implement an LSTM-based RNN model for handling sequential text data.
• Apply proper text preprocessing and transformation techniques.
• Evaluate model performance using metrics like accuracy.
• Create an intuitive sentiment score and emoji output for user-friendliness.

---

4. Model Architecture

• Embedding Layer: Converts word indices into dense vector representations.
• Bidirectional LSTM Layer: Captures contextual dependencies in both forward and backward directions.
• Dense Output Layer: Uses sigmoid activation for binary classification.

---

5. Implementation Steps

1. Load IMDB Dataset: Preprocessed movie reviews with word indices.
2. Preprocess Data: Pad sequences to a fixed length for uniform input.
3. Build and Train Model: Implement Bidirectional LSTM with embedding layer.
4. Evaluate Performance: Test the model with validation data.
5. Custom Predictions: Encode new reviews, predict sentiment, and classify into positive, negative, or neutral categories.

---

6. Requirements

To run the project, install the required dependencies:

pip install tensorflow numpy pandas

---

7. How to Run

1. Train the model: Run the Python script to train and save the model.
2. Predict Sentiment: Load the saved model and run it on custom reviews.
3. Review Output: Sentiment scores and emoji representation.

---

8. Results & Performance

• Evaluation Metric: Binary accuracy
• Accuracy Achieved: ~85% (varies with training settings)
• Sentiment Categories:
  • Positive (>65% score) 😄
  • Neutral (35%-65% score) 😐
  • Negative (<35% score) 😞

---

9. Conclusion

This project successfully demonstrates sentiment analysis using RNN with LSTM on IMDB reviews. The model captures text sentiment effectively and provides a user-friendly interpretation with sentiment scores and emojis.

---

10. Future Improvements

• Fine-tuning hyperparameters for better accuracy.
• Using a pre-trained word embedding like GloVe.
• Expanding to multi-class sentiment analysis.
• Deploying as a web application using Flask or Streamlit.

---

11. Authors

• Abhijeet
• Jhalki Kulshrestha

Code - Abhijeet Project Report - Abhijeet & Jhalki