Towards improving community detection in multilayer networks using semi-supervised matrix factorization

doi:10.1007/s10660-025-10020-4

Towards improving community detection in multilayer networks using semi-supervised matrix factorization

2025-07-17

0

PRE

AI

DOI：10.1007/s10660-025-10020-4

Original

Original request for help

Save

Abstract

En 中文

Community detection in multilayer networks faces significant challenges due to the neglect of user attribute information and the inability to account for the heterogeneity among layers. These limitations often result in suboptimal performance and less meaningful community structures. To overcome these challenges, this paper introduces a novel approach for community detection in multilayer networks using the Semi-Supervised Matrix Factorization Algorithm (SSMFA). Our method integrates structural, content-based, and user overlap information into a unified nonnegative matrix factorization framework, ensuring a holistic representation of the multilayer network. Community structures for individual layers are independently identified through a semi-supervised clustering process that leverages pairwise constraints to improve clustering precision. Subsequently, ensemble clustering is applied to merge the detected community structures into a cohesive and globally optimized network representation. Extensive experimental evaluations demonstrate that SSMFA significantly outperforms existing state-of-the-art techniques across multiple evaluation metrics, highlighting its effectiveness in capturing intricate community structures in multilayer networks. Specifically, on the Tumblr network, our method achieves a 1.2% improvement in modularity compared to the best-performing baseline.

Keywords:

Community detection

Multilayer network

Matrix factorization

Semi-supervised learning

Towards improving community detection in multilayer networks using semi-supervised matrix factorization

Abstract

Journal

Organization