Tweet Emotion Recognition with TensorFlow

🚀 Overview

This project implements a deep learning model using TensorFlow to classify emotions expressed in tweets. It is based on a guided Coursera project and leverages the dair-ai/emotion dataset available on Hugging Face.

The model is trained to recognize six emotion categories in tweets:

• Joy
• Anger
• Sadness
• Love
• Fear
• Surprise

The project includes text preprocessing, embedding, LSTM model training, and performance evaluation.

🎯 Features

• 🔍 Emotion classification from tweet text
• 🧠 LSTM-based deep learning model
• 💬 Tokenization, padding, and word embeddings
• 📊 Confusion matrix and classification report

🏗️ Project Structure

├── Tweet_Emotion_Recognition.ipynb  # Main Jupyter notebook
├── Tweet_Emotion_Recognition.pdf    # PDF export of the notebook
├── requirements.txt                 # Python dependencies
└── README.md                        # Project documentation

🔥 Technologies Used

Category	Tools & Libraries
Deep Learning	TensorFlow, Keras
NLP	Hugging Face Datasets
Visualization	Matplotlib
Data Handling	NumPy
Development	Jupyter Notebook

Requirements

• Python 3.8+
• TensorFlow
• NumPy
• Matplotlib
• Scikit-learn
• Hugging Face Datasets
• Jupyter

Install required packages using:

pip install tensorflow numpy matplotlib scikit-learn datasets jupyter

How to Run

1. Clone the repository:

   git clone https://github.com/your-username/tweet-emotion-recognition.git
   cd tweet-emotion-recognition

2. (Optional) Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Open and run the notebook:

   jupyter notebook Tweet_Emotion_Recognition.ipynb

Model Architecture

The model is a Sequential neural network composed of:

• An Embedding layer (using pre-trained GloVe vectors)
• A Bidirectional LSTM layer with 64 units (returning sequences)
• A second Bidirectional LSTM layer with 32 units
• A Dense layer with ReLU activation
• A Softmax output layer for emotion classification

Results

The model was trained and evaluated on the dair-ai/emotion dataset and achieved:

• Validation Accuracy: ~89.5%
• Test Accuracy: 88.5%
• F1 Score: 88.5%

Results were visualized using a confusion matrix and classification report.

License

This project is for educational purposes and is based on a guided Coursera project: Coursera: Tweet Emotion Recognition with TensorFlow

Author

Khaled Jamal
LinkedIn