🌍 Multilingual-Multilabel Sentiment Analysis

A deep learning project for analyzing multilingual text sentiments (Tamil and Tulu) using transformers and graph-based approaches.

---

✨ Features

• 🔄 Fine-tuning XLM-RoBERTa for multilingual sentiment classification.
• 🌐 Graph Neural Networks (GNN) for sentiment analysis.
• 🎭 Support for multiple sentiment classes: Positive, Negative, Neutral, Mixed Feelings, etc.
• ⏱️ Training with Early Stopping and Model Checkpointing.
• 📊 F1-Score Based Evaluation for robust performance.

---

⚙️ Installation

🐍 Using Conda

1. Create a new conda environment:

   conda create -n sentiment python=3.9  
   conda activate sentiment  

2. Install basic dependencies:

   conda install pytorch torchvision -c pytorch  
   pip install -r requirements.txt  

3. For graph-based approaches, install additional dependencies:

   pip install -r graph_requirements.txt  

📦 Using pip only

Install all required packages:

pip install -r requirements.txt  
pip install -r graph_requirements.txt  # Optional: for graph-based approach  

---

📂 Dataset Structure

Ensure your dataset files are in the dataset/ folder:

• cleaned_tamil_dev.csv
• cleaned_tamil_train.csv
• Tam-SA-train.csv
• Tam-SA-val.csv
• Tulu_SA_train.csv
• Tulu_SA_val.csv

---

🚀 Usage

Training the Models

For transformer-based approach:

python main.py  

For graph-based approach:

python graph_approach.py  

Testing a Model

python test_model.py  

💡 Use this command to test the trained model on new text.

---

🛠️ Model Architectures

• Transformer Approach: Fine-tuned XLM-RoBERTa base model.
• Graph Approach: Graph Convolutional Networks (GCN) with custom tokenization.

---

📈 Performance

• 🤖 Transformer Model: ~0.63 F1 score.
• 🌐 Graph Model: Comparable performance with unique strengths.

---

📁 Project Structure

.  
├── dataset/                  # Data files  
├── main.py                  # Transformer-based implementation  
├── graph_approach.py        # GNN-based implementation  
├── test_model.py            # Model inference code
├── FInetuning_ XLM-RoBERTa.ipynb         #Jupyter notebook
├── requirements.txt         # Basic dependencies  
└── graph_requirements.txt   # Additional GNN dependencies 

---

💻 Hardware Used

• NVIDIA 4060 GPU with 8GB RAM.
• CUDA 12.1 and PyTorch 2.5.1.
• 🖥️ Tested on Google Colab T4 GPU (15GB GPU RAM).