Skip to content

Repository files navigation

Graph Representation Learning using Feature Fusion

This project explores a simple yet effective approach for graph learning by combining multiple graph and node features into a single feature vector. Instead of relying only on Graph Neural Networks (GNNs), the project evaluates a Multi-Layer Perceptron (MLP) as a lightweight alternative across multiple benchmark datasets.

Features

  • Graph feature fusion through feature concatenation
  • MLP-based graph classification
  • Evaluation on multiple graph datasets
  • Performance comparison across datasets
  • Modular and easy-to-extend implementation

Technologies

  • Python
  • PyTorch
  • NumPy
  • Scikit-learn
  • Jupyter Notebook

Workflow

  1. Load graph dataset.
  2. Extract multiple graph features.
  3. Concatenate features into a unified representation.
  4. Train an MLP classifier.
  5. Evaluate model performance.

Goal

Investigate whether a simple feature engineering approach combined with an MLP can achieve competitive performance on graph learning tasks while reducing model complexity.

Author

Parisa Arbab

Create Environment

It is better to create a virtual environment for this project.

python3 -m venv <myenvpath>

Install all required packages

pip install -r requirements.txt

About

Developed a graph representation learning framework that combines multiple structural and node features into one feature vector for classification tasks. The project evaluates an MLP-based approach as a lightweight alternative to Graph Neural Networks across multiple benchmark datasets, comparing performance, scalability, and feature effectiveness.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages