IMDB Movie Review Sentiment Analysis

This project performs sentiment analysis on the IMDB movie reviews dataset using Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks. It includes a production-ready Flask web application with a TailwindCSS frontend to predict whether movie reviews are positive or negative, complete with a user-friendly interface and search history feature.

---

Problem Statement

Given a movie review (text), predict whether the review is positive or negative. This is a binary classification problem in Natural Language Processing (NLP).

Dataset Link: IMDB Dataset of 50K Movie Reviews

Kaggle Notebook Link: Movie Sentiment Analysis with LSTM-RNN

Sentiment Analyser Overview

---

Features

• Model: Pre-trained LSTM model for accurate sentiment prediction on movie reviews.
• API: Flask-based RESTful API to serve predictions, supporting single and batch review inputs.
• Frontend: Responsive web interface built with TailwindCSS and Chart.js, featuring:
  • Input textarea for single or multiple reviews.
  • Sentiment results with confidence scores and color-coded badges (green for positive, red for negative).
  • Pie chart for batch review sentiment distribution.
  • Search history to track all predictions within a session, with a clear button (resets on refresh).
• Tech Stack:
  • Backend: Flask, TensorFlow, NLTK, NumPy
  • Frontend: HTML, TailwindCSS, Chart.js, JavaScript

---

Concepts Used

1. Word Embedding & Feature Representation

• Raw text is converted into numeric vectors using word embeddings.
• Embeddings map words into a continuous vector space, preserving semantic relationships.
  • Example: king → [0.25, -0.63, 0.41, ...]
  • Words with similar meanings have closer embeddings.
• Implemented using Keras Embedding layer.

2. Recurrent Neural Network (RNN)

• RNNs process sequential data like text, capturing dependencies between words.
• Example: In "The movie was not good," the word "not" influences the sentiment.
• Limitation: Suffers from vanishing gradients, struggling with long-range dependencies.

3. Long Short-Term Memory (LSTM)

• An advanced RNN variant with gates (input, forget, output) to control memory flow.
• Benefits:
  • Remembers important context over long sequences.
  • Ignores irrelevant words.
  • Mitigates vanishing gradient issues.
• For sentiment analysis, LSTM captures context like "not" affecting "good" in long reviews.

4. Web Application

• Flask API: Serves the pre-trained LSTM model, handling text preprocessing and predictions.
• Frontend: Responsive UI with TailwindCSS for styling, Chart.js for visualizations, and JavaScript for interactivity.
• History: Client-side storage of search history (resets on refresh) with a clear button.

---

Project Workflow

1. Data Preparation

   • Loaded IMDB dataset (50,000 reviews).
   • Tokenized text and padded sequences to a fixed length (maxlen=250).
   • Split into train/test sets.

2. Model Building

   • Built RNN and LSTM models with an Embedding layer.
   • Used early stopping to prevent overfitting.
   • Saved the trained LSTM model (model.h5) and tokenizer (tokenizer.pickle).

3. API Development

   • Created Flask endpoints:
     • /: Serves the main web interface.
     • /predict: Processes single or batch reviews, returns sentiment and confidence scores.
   • Preprocessing includes tokenization, stopword removal, stemming, and padding.

4. Frontend Development

   • Designed a responsive UI with TailwindCSS.
   • Added interactive elements: input form, results display, batch summary chart, and search history.
   • Styled with Poppins font and vibrant colors (indigo, green, red).

5. Training

   • Optimizer: Adam
   • Loss: Binary Crossentropy
   • Metrics: Accuracy

---

Model Architecture

Simple RNN:

model = Sequential()
model.add(Embedding(10000, 128))
model.add(SimpleRNN(128, activation='relu'))
model.add(Dense(1, activation='sigmoid'))

LSTM (Improved):

model = Sequential()
model.add(Embedding(10000, 128))
model.add(LSTM(128))
model.add(Dense(1, activation='sigmoid'))

---

Results

• Simple RNN:

  • Test Accuracy: 0.86
  • Test Loss: 0.32
  • Struggled with long reviews due to vanishing gradients.

• LSTM:

  • Test Accuracy: 0.88
  • Test Loss: 0.28
  • Improved performance by capturing long-term dependencies.

• Web App:

  • Successfully predicts sentiments for single or batch reviews.
  • Example output:

    Review: "This movie was amazing with great acting and story!"
    Sentiment: Positive, Confidence: 0.87
    Review: "The moview is utter shit. It is not upto the mark."
    Sentiment: Negative, Confidence: 0.12
    

---

How to Run

Prerequisites

• Python: 3.8 or higher
• Files:
  • app.py: Flask application
  • templates/index.html: Frontend template
  • lstm_model.h5: Pre-trained LSTM model
  • tokenizer.pickle: Saved tokenizer
• Internet: Required for TailwindCSS and Chart.js CDNs (or host locally for offline use)

Directory Structure

project_directory/
├── app.py
├── models/
    └── lstm_model.h5
    └── tokenizer.pickle
├── templates/
│   └── index.html

If lstm_model.h5 is missing do run the Notebook and save the model

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/BenGJ10/Movie-Sentiment-Analysis.git
   cd Movie-Sentiment-Analysis

2. Create a Virtual Environment:

   python -m venv venv

   Activate it:

   • Windows:

     venv\Scripts\activate

   • macOS/Linux:

     source venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Run the App:

   python app.py

   • Access at http://127.0.0.1:5000 in your browser.
   • The app will load the LSTM model and tokenizer at startup.

5. Usage:

   • Enter one or multiple reviews (one per line) in the textarea.
   • Click Analyze Sentiment to see results with confidence scores.
   • View batch sentiment distribution (pie chart) for multiple reviews.
   • Check the Search History section for past searches.
   • Click Clear History to reset results and history (resets automatically on refresh).

---

References

• IMDB Sentiment Dataset
• Understanding LSTMs
• Keras Embedding Layer Docs

---

Contributing

Contributions are welcome! Please open an issue or submit a pull request on GitHub.

Author

Developed by Ben Gregory John