Customer Churn

Harnessing state-of-the-art NLP and deep learning methods to decode emotions embedded within movie critiques.

Sentiment Dashboard

Project Overview

This project leverages Natural Language Processing (NLP) and Deep Learning techniques to analyze and predict sentiment in movie reviews. By integrating multiple text preprocessing strategies and advanced neural network architectures, the system identifies whether a review expresses positive or negative sentiment. The project further explores the use of Transfer Learning and pre-trained embeddings to enhance prediction accuracy and generalization.

Dataset

The dataset was sourced from Kaggle and is based on IMDb movie reviews. It contains 50,000 reviews labeled as either positive or negative sentiments and was split into training and testing sets for model development. Kaggle Dataset Link

Key Visualizations

Sentiment Distribution

Sentiment Distribution

Shows the count of positive and negative sentiments in movie reviews, illustrating dataset balance.

Confusion Matrix

Confusion Matrix

Visualizes the model’s prediction accuracy by comparing actual vs predicted sentiment labels.

Model Accuracy over Epochs

Model Accuracy Over Epochs

Displays training and validation accuracy progression across epochs for the LSTM sentiment model.

Observation & Conclusion

Deep learning models, particularly the LSTM with pre-trained GloVe embeddings, outperformed traditional classifiers in sentiment prediction on movie reviews. Data preprocessing and embedding layers significantly improved model accuracy, with the LSTM achieving over 90% accuracy. However, simpler models like CNN and simple neural networks also showed competitive results. This demonstrates the importance of combining effective text preprocessing with advanced model architectures for sentiment analysis.

Project Links

GitHub Repository: GitHub

Real-World Impact

This project enables accurate sentiment classification in movie reviews, aiding platforms in understanding audience feedback and improving recommendations. The system’s ability to predict sentiment and probable ratings in real-time can be extended to customer sentiment monitoring, brand reputation management, and automated review analysis across domains such as entertainment, retail, and social media.