Community Detection via Structured Adaptive Block-Diagonal Learning with Topology-Subspace Fusion

Abstract

Subspace clustering can capture the structural features of complex networks. However, existing subspace-based methods encounter challenges related to dependence on priori knowledge, difficulties in exploiting the connection between two stages and addressing networks with fuzzy boundaries. In response to these challenges, a novel subspace clustering-based community detection algorithm, Structured Adaptive Block Diagonal Subspace Learning with Fusion (SABDSLF) is proposed. First, adaptive block diagonal subspace learning strategy is designed. This approach establishes a convex objective function without the need for prior knowledge. Second, structured subspace learning strategy uses a structure matrix to capture the connection between the two stages of the subspace algorithm. Finally, an information fusion strategy is designed to combine topological information and subspace information, enabling the handling of complex networks with fuzzy boundaries. Experiments were conducted on real-world and synthetic networks, demonstrating that SABDSLF outperforms several state-of-the-art community detection methods in terms of precision and robustness.