Multi-Task & Emotion Analysis (don't give a fck, it's shit (used ai))

This project builds a multi-task learning model to classify text for three tasks:

1. Aspect-Based Sentiment Analysis (positive/negative/neutral) 💡
2. Emotion Detection (happy/sad/angry) 😄😢😡
3. Sentiment-Emotion Joint Modeling (combining sentiment & emotion)

📋 Requirements

• Python 3.6+
• TensorFlow (Keras) & PyTorch
• Libraries: pandas, numpy, nltk, spacy, scikit-learn, etc.

Install all dependencies with:

pip install -r requirements.txt

🚀 Setup & Installation

1. Clone the repo:

   https://github.com/JeetChaurasia/Multi-Task-Learning-MTL-Model.git

2. Download the spaCy model:

   python -m spacy download en_core_web_sm

3. Get the Semeval-2014 dataset and save as semeval2014_train.csv & semeval2014_test.csv.

🧠 Model Architecture

• Input: Preprocessed text (tokenized, stopword removed)
• Embedding Layer: Converts tokens to dense vectors
• LSTM Layer: Captures text sequence dependencies
• Outputs:
  • Aspect Sentiment (positive/negative/neutral)
  • Emotion (happy/sad/angry)

Adversarial training improves model robustness using PyTorch.

🎓 Training

Train the model with:

python train.py

⚡ Metrics:

• Accuracy 📊
• F1 Score 🔥
• ROC AUC 🎯
• MSE 📉

📝 Save/Load Model

Save the model with:

model.save('multi_task_learning_model.h5')

Load the model:

loaded_model = keras.models.load_model('multi_task_learning_model.h5')

📄 License

MIT License - see LICENSE for details.