IMDB-Sentiment-Classification-using-BERT

🧠 Project Overview

This project focuses on building a sentiment classification model using the IMDB movie review dataset.

✅ Key Highlights

• 📊 Processed and cleaned 75,000 movie reviews to ensure high-quality input for training.
• 🔍 Fine-tuned a Small BERT model with custom encoder layers and an optimized classification head.
• ⚙️ Enhanced performance using AdamW optimizer, learning rate tuning, and weight decay to improve generalization.
• 📈 Achieved 72% training accuracy and 75% validation accuracy, demonstrating effective model fine-tuning and deep learning optimization.

The project showcases the power of transfer learning in NLP using BERT and TensorFlow.

This project demonstrates fine-tuning a BERT model on the Stanford IMDB movie review dataset for binary sentiment classification (positive/negative). The model is built using TensorFlow, TensorFlow Hub, and TensorFlow Text.

🚀 Features

• ✅ Downloads and preprocesses the IMDB dataset
• ✅ Fine-tunes a Small BERT model (bert_en_uncased_L-4_H-512_A-8) from TensorFlow Hub
• ✅ Splits data into training, validation, and test sets
• ✅ Uses EarlyStopping and ModelCheckpoint callbacks
• ✅ Saves the model in .h5, .keras, and TensorFlow SavedModel formats

📦 Libraries Used

• TensorFlow
• TensorFlow Hub
• TensorFlow Text
• Keras

📁 Dataset Structure

• Training: pos, neg, and unsup
• Test: pos, neg
• Total files: ~100,000

🧠 Model Architecture

• BERT encoder (trainable)
• Dense(64) + Dropout(0.3)
• Final sigmoid output for binary classification

🏁 How to Run

pip install tf-models-official tensorflow tensorflow_hub tensorflow_text