This repository implements a hybrid approach to hypergraph community detection by combining embedding-based clustering with H-Louvain algorithm. The goal is to improve modularity optimization in hypergraphs through a two-stage process.
- Hypergraph Embedding: Transform the hypergraph structure into a vector space representation
- Initial Clustering: Apply clustering algorithms to the embeddings to identify initial communities
- H-Louvain Refinement: Use the initial clusters as starting points for H-Louvain algorithm
- Performance Evaluation: Compare results with standalone EC-Louvain and H-Louvain approaches
We implement various embedding techniques for hypergraphs, which capture the higher-order relationships between nodes.
After obtaining node embeddings, we apply clustering algorithms to identify initial community structures.
Instead of random initialization, we use the clusters obtained from embedding-based methods as the starting point for H-Louvain.
- Improved modularity scores compared to standalone methods
- Better community detection in complex hypergraphs
- Potentially faster convergence of the H-Louvain algorithm
This section will be updated with experimental results and performance comparisons.
- EC-Louvain algorithm
- H-Louvain algorithm
- Hypergraph embedding techniques
The following repositories are used as baselines for our implementation:
- H-Louvain - Implementation of the Hypergraph Louvain algorithm
- ECCD - Edge-Centric Community Detection (EC-Louvain)
- OpenNE - Open Network Embedding framework for node representation learning
Additional repositories may be incorporated as the project evolves. The implementation also uses standard Python and Julia libraries not listed here.
This work is inspired by:
- EC-Louvain/Leiden: Pankratz et al., 2024
- h-Louvain (Hypergraph Modularity): Kamiński et al., 2024
- Equivariant Hypergraph Diffusion Neural Operators: Huang et al., 2022