Multi-View Spectral Clustering for Graphs with Multiple View Structures

Overview

GenClus is a graph clustering method aimed at modeling multi-view graphs where each view may correspond to a different clustering of nodes. Its design is built on principled foundations, generalizing various spectral clustering methods while also facilitating the derivation of computationally efficient numerical solutions.

This repo contains an implementation of GenClus along with the complete experiment scripts of the associated paper

Yorgos Tsitsikas and Evangelos E. Papalexakis. Multi-View Spectral Clustering for Graphs with Multiple View Structures. In Proceedings of the 2025 SIAM International Conference on Data Mining (SDM), 2025

The full version of this paper can be accessed at https://arxiv.org/abs/2501.11422

Also, BibTeX users can cite this work as follows:

@inproceedings{genclus,
  title={Multi-View Spectral Clustering for Graphs with Multiple View Structures},
  author={Tsitsikas, Yorgos and Papalexakis, Evangelos E.},
  booktitle={Proceedings of the 2025 SIAM International Conference on Data Mining (SDM)},
  year={2025},
}

Dependencies

genclus.m

• MATLAB - Required (Tested on MATLAB 2019b)
• MTIMESX - Optional

clustering_quality_on_artificial_data.m

• MATLAB - Required (Tested on MATLAB 2019b)
• Multi-Graph Explorer - Required

real_world_case_study.m

• Openflights data:
  • airlines.dat, airports.dat, countries.dat, routes.dat - Required
• Continents data:
  • continents2.csv - Required
• MATLAB - Required (Tested on MATLAB 2019b; Currently fails in MATLAB 2022b)
• Multi-Graph Explorer - Required
• Tensor Toolbox - Required

execution_time_on_artificial_data.m

• MATLAB - Required (Tested on MATLAB 2019b)
• Multi-Graph Explorer - Required

License

3-Clause BSD