Skip to content

README.md

Latest commit

 

History

History
102 lines (76 loc) · 3.66 KB

README.md

File metadata and controls

102 lines (76 loc) · 3.66 KB
Raw

🎭 Text Emotion Classifier

An end-to-end Machine Learning and Streamlit application that classifies the underlying emotion of any input English text into one of six categories: Sadness (0), Joy (1), Love (2), Anger (3), Fear (4), and Surprise (5).

This project leverages a large dataset containing over 416,000 text samples to compare classical Machine Learning approaches (Logistic Regression, Linear Support Vector Machines, Random Forests) to achieve a considerably decent accuracy and a deployable solution.

🚀 Live Demo & Deployment

The application is designed to run via a Streamlit dashboard.

⚡ Quick Preview (via Streamlit)

You can view this on know-your-emotions.streamlit.app and test it. Thanks, Streamlit, for helping us deploy the app!

📊 Dataset

The model is trained on the popular Emotions Dataset by Nelgiriyewithana via Kaggle. It contains highly unbalanced categories:

  • Joy (~141k samples)
  • Sadness (~121k samples)
  • Anger (~57k samples)
  • Fear (~47k samples)
  • Love (~34k samples)
  • Surprise (~15k samples)

🛠️ Model Architecture & Evaluation

We benchmarked multiple representation techniques and classifiers:

  1. TF-IDF + Logistic Regression
    • Unigrams vs Bigrams with sublinear_tf and custom class weighting.
    • Accuracy: ~90-91%
  2. TF-IDF + Linear SVC (Best Performing Classical Model)
    • Achieved solid macro performance with balanced weights.
    • Accuracy: ~90% (with exceptional speed and small disk footprint)
  3. Sentence Transformers (all-MiniLM-L6-v2) + Logistic Regression
    • Evaluated dense semantic embeddings.
    • Accuracy: ~65%
  4. TF-IDF + Random Forest Classifier
    • Constrained to prevent unconstrained tree size.
    • Accuracy: ~81%

Our chosen production pipeline employs the Linear Support Vector Classifier (LinearSVC) paired with a tuned Bigram TF-IDF Vectorizer due to its outstanding generalization and lightweight inference footprint.

📂 Repository Structure

├── app.py                  # Streamlit web application script
├── emotion_model.pkl       # Saved trained LinearSVC classifier
├── tfidf_vectorizer.pkl    # Saved fitted TF-IDF Bigram Vectorizer
├── requirements.txt        # Package dependencies
└── README.md               # Project documentation

⚙️ Installation & Local Setup

Follow these steps to run the Streamlit app on your local machine:

1. Clone the repository

git clone https://github.com/githubsneethi/Text-Emotion-Classifier.git
cd Text-Emotion-Classifier

2. Create and activate a virtual environment

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

3. Install dependencies

Ensure you have python installed, then run:

pip install streamlit scikit-learn numpy pandas joblib

4. Run the Streamlit App

streamlit run app.py

Your browser should automatically pop open a new window to http://localhost:8501 showcasing the interface.

🧪 Streamlit App Features

  • Text input area: Write any phrase or paragraph of your choice.
  • Dynamic Emotion prediction: Uses the cached serialized model pipeline instantly.
  • Decision-function Softmax: Outputs confidence probabilities for each class (even with LinearSVC).
  • Clean Bar Charts: Sorted breakdowns of the exact classification confidence percentages across all emotions.

🤝 Contribution

Contributions, issues, and feature requests are welcome! Feel free to check the issues page or open a Pull Request.

📄 License

This project is licensed under the MIT License.