📚 Comparative Analysis of Graph Neural Network and Sequence Models for News Summarization

This project presents a comparative study of four different neural network architectures for text summarization:

• Transformer
• BiLSTM
• GNN + Transformer
• GNN + BiLSTM

The goal is to explore how Graph Neural Networks (GNNs) can enhance traditional sequence models by capturing document structure beyond simple sequential relationships. The project uses the CNN/DailyMail dataset and evaluates the models using ROUGE metrics.

---

✨ Key Features

• Implements Transformer-based, BiLSTM, and GNN-enhanced hybrid models.
• Builds heterogeneous document graphs with words and sentences as nodes.
• Uses Graph Attention Networks (GATs) for message passing between graph nodes.
• Integrates GNN representations with BART Transformer and BiLSTM.
• Includes preprocessing pipeline: cleaning, tokenization, graph construction.
• Evaluation using ROUGE-1, ROUGE-2, and ROUGE-L metrics.
• Reproducible notebooks: GNN+Transformer.ipynb, GNN+BiLSTMs.ipynb, Transformer.ipynb, bilstm.ipynb.

---

⚙️ Tech Stack

• Language: Python
• Frameworks/Libraries:
  • PyTorch
  • PyTorch Geometric
  • Transformers (Hugging Face)
  • NLTK
• Models: BART, BiLSTM, Graph Attention Networks
• Dataset: CNN/DailyMail

---