Community Detection Experiment

Overview

This repository presents an experimental implementation of community detection algorithms tailored for hypergraphs. It explores a novel approach by transforming hyperedges into a two-section graph using node embedding techniques (Node2Vec and DeepWalk) followed by initial partitioning with K-Means. The resulting graph is then analyzed using three prominent community detection algorithms: Louvain, Leiden, and a hierarchical Louvain variant (H-Louvain).

Key Features

• Hypergraph to Two-Section Graph Conversion: Scripts to generate a two-section graph representation from hyperedge data using Node2Vec and DeepWalk embeddings.

• Initial Partitioning via K-Means: Implementation of K-Means clustering to obtain initial community assignments for the two-section graph (assumes k=10, easily adjustable).

• Community Detection Algorithms:

  • Implementation of the Louvain algorithm.
  • Implementation of the Leiden algorithm.
  • Implementation of the H-Louvain algorithm applied to the generated two-section graph.
  • Pure implementation of the H-Louvain algorithm for direct hypergraph analysis (based on the pawelwm/h-louvain repository).

• Evaluation Metrics: Automatic calculation and recording of Modularity and Adjusted Mutual Information (AMI) for evaluating the detected communities.

• Dataset Integration: Utilizes datasets provided in the pawelwm/h-louvain repository for experimentation.

Potential Future Directions (Optional)

• Experiment with different node embedding techniques and parameters.
• Explore alternative initial partitioning methods.
• Implement additional hypergraph community detection algorithms.
• Develop visualization tools for hypergraphs and their community structures.
• Evaluate performance on a wider range of hypergraph datasets.

Contributing

Contributions are welcome! Please feel free to submit pull requests or contact the maintainers directly to discuss potential contributions.

License

This project is licensed under the terms of the MIT License.

Acknowledgements

• This work utilizes datasets and the pure H-Louvain implementation from the H-Louvain repository. The H-Louvain algorithm and hypergraph modularity are described in: Kamiński et al., 2024.

• The project may also incorporate concepts or algorithms from the ECCD repository. The EC-Louvain/Leiden algorithm is described in: Pankratz et al., 2024.