Attributed Graph Clustering via Coarsening with Modularity

Samarth Bhatia, Yukti Makhija, Manoj Kumar, Mohit Kataria, Sandeep Kumar

Primary Area: learning on graphs and other geometries & topologies

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Graph Clustering, Graph Neural Networks, Convex Optimization, Non-Convex Optimization, Graph Coarsening

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A novel unsupervised framework incorporating modularity with important graph coarsening regularization terms to improve clustering (via coarsening).

Abstract: Graph clustering is a widely used technique for partitioning graphs, community detection, and other tasks. Recent graph clustering algorithms depend on combinations of the features and adjacency matrix, or solely on the adjacency matrix. However, in order to achieve high-quality clustering, it is necessary to consider all these components. In this paper, we propose a novel unsupervised learning framework that incorporates modularity with graph coarsening techniques and important graph regularization terms that improve the clustering performance. Furthermore, we also take into account Dirichlet energies for smoothness of signals, spectral similarity, and coarsening reconstructional error. The proposed framework is solved efficiently by leveraging block majorization-minimization, $\log\det$ of the Laplacian, smoothness and modularity, and is readily integrable with deep learning architectures such as GCNs and VGAEs in the form of losses. Extensive theoretical analysis and experiments with benchmark datasets elucidate the proposed framework’s efficacy in graph clustering over existing state-of-the-art methods on both attributed and non-attributed graphs.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9264