Citation Network Link Prediction Using Graph Neural Networks

Project Description

This project focuses on predicting citation links in academic papers using Graph Neural Networks (GNNs). By constructing and analyzing a citation network where nodes represent papers and edges represent citations, the model predicts potential future citations by leveraging node features and the graph structure.

Technologies Used

• Programming Language: Python
• Frameworks and Libraries:
  • PyTorch
  • PyTorch Geometric
  • NetworkX (for graph manipulation)
  • Plotly (for interactive 3D visualization)
  • Scikit-learn (for data preprocessing and evaluation)
• Graph Neural Network Models:
  • Graph Convolutional Network (GCN)
  • GraphSAGE

Key Features

• Graph Neural Network Architecture:
  • Designed and implemented GCN and GraphSAGE models for link prediction.
• Data Preprocessing:
  • Processed the Cora citation dataset, extracted features, and constructed the adjacency matrix.
• Custom Functions:
  • Developed custom functions to handle dynamic graph structures for link prediction.
• Model Training and Optimization:
  • Conducted model training and hyperparameter tuning to optimize performance.
• Visualization:
  • Visualized the citation network and prediction results using interactive 3D plots.

How It Works

1. Data Preparation:
   • The Cora citation dataset is preprocessed to create a feature matrix and adjacency matrix representing the citation network.
2. Model Implementation:
   • Implemented GCN and GraphSAGE architectures using PyTorch and PyTorch Geometric.
3. Training and Evaluation:
   • Trained the models on the preprocessed dataset and evaluated their performance on link prediction tasks.
4. Visualization:
   • Used NetworkX and Plotly to create interactive 3D visualizations of the citation network and prediction results.

Usage

Since this project is contained within a Jupyter Notebook (.ipynb), you can directly open and run the notebook to explore the data preprocessing steps, model implementation, training, and visualization.

Results and Impact

• Accuracy:
  • Achieved high accuracy in predicting citation links, demonstrating the effectiveness of GNNs for link prediction tasks.
• Applications:
  • The model's predictions provide insights into potential future citations, which can be valuable for academic recommendation systems and research impact analysis.

Author

• Anurag Upperwal:
• Email: anuragupperwal@gmail.com

Acknowledgments

• This project uses the Cora citation dataset.
• The implementation leverages the PyTorch Geometric library for GNN operations.