Graph Clustering Based on Frobenius Norm Low-Rank Approximation

Graph clustering is a technique that used to uncover the structural patterns in the graph and reduce its dimension. Multiple algorithms have been developed to approach the task of graph clustering. Despite the progress that has been done, there are still many challenges in graph clustering such as clustering noisy graphs. In this paper, we introduce a graph clustering algorithm that approximates the graph's similarity matrix into a lower-rank one while conserving its fundamental structure and eliminating noise and redundancy. The approximated lower-rank matrix is then used to reveal the clusters through symmetric non-negative matrix factorization. The problem is solved through an alternating minimization framework. The conducted experiments on real-world graphs show that the proposed model enables efficient and accurate clustering results better than other existing algorithms.