Jointly subspace clustering and manifold learning for attributed graphs

Khodaparast, Elham; Jamshidi, Mina

doi:10.22124/jmm.2026.33369.3050

Jointly subspace clustering and manifold learning for attributed graphs

Articles in Press

Document Type : Research Article

Authors

Elham Khodaparast ¹
Mina Jamshidi ²

¹ Department of Applied Mathematics, Graduate University of Advanced Technology, Kerman, Iran

² Department of Applied Mathematics, Graduate university of Advanced Technology, Kerman, Iran

10.22124/jmm.2026.33369.3050

Abstract

Attributed graph clustering, or community detection, groups data with predefined graph similarity matrix based on both topological structure of the graph and node features. The features of nodes and their connections may not be fully consistent, leading to a potential mismatch between node attributes and graph structure in attributed graph clustering. Therefore, one of the main goals of attributed graph clustering is to learn similarity and feature representations that are as compatible as possible. Continuing this goal, in this paper we provide a method in which a new similarity matrix is constructed as close as possible to the predefined similarity matrix. Moreover, it simultaneously captures the subspace structure of data points and preserves the manifold structure induced by the feature matrix which causes more smoothness in the new graph structure. Results on real-world datasets demonstrate the effectiveness of our method in community detection.

Keywords

Main Subjects