ML Twitter Sentiment Analysis

This project explores various deep learning approaches to sentiment classification using the dair-ai/emotion dataset. The goal is to predict emotional tone (e.g., joy, sadness, anger) from tweets using different neural network architectures.

📊 Dataset

The dataset consists of 20,000 labeled text samples split into training, validation, and test sets. Each entry includes:

• text: A string of user-generated content (similar to tweets)
• label: An emotion category:
  • 0 – Sadness
  • 1 – Joy
  • 2 – Love
  • 3 – Anger
  • 4 – Fear
  • 5 – Surprise

---

🧠 Models Implemented

✅ Baseline Models:

• Simple RNN: Single and stacked layers to process sequential text data
• LSTM (Long Short-Term Memory): Better captures long-term dependencies in sequences

✅ Advanced Models:

• BERT (via TensorFlow Hub): Pre-trained transformer for contextual embedding and classification
• Cohere Classification API: Integration of Cohere's NLP models via API

---

🔧 Key Concepts & Techniques

• Tokenization and padding of input sequences
• Text sequence length distribution analysis
• Model performance visualization (accuracy/loss over epochs)
• TensorFlow's Keras API for building and training models
• Comparative analysis of RNN, LSTM, and Transformer-based models
• Integration with APIs (Cohere)

---

📚 Educational Value

This notebook:

• Demonstrates the evolution of NLP models from RNNs to Transformers
• Showcase best practices for evaluating sequence models
• Introduce pre-trained models and modern NLP APIs