Community Detection Using Nonnegative Matrix Factorization and Sparse Linear Coding

ABSTRACT Community detection plays a fundamental role in uncovering hidden structures within complex networks. It simplifies a better understanding of relationships and interactions among nodes. In this article, we propose a novel approach for community detection using a sparse nonnegative matrix factorization (NMF) algorithm by leveraging sparse linear coding (SLC). Most of the community detection algorithms are based on graph‐theoretic analysis, which makes it difficult to identify precise community boundaries when the difference between intra‐community connections and inter‐community connections is quite less. To address this particular issue, we use SLC to identify the subspace using the condition of dependence. Further, we use NMF for clustering as well as to overcome the problem of overlapping. The positive aspects of this approach include its interpretability, feature extraction, soft partitioning solutions, and easy framework. The proposed approach has been evaluated on a combination of real‐world networks and synthetic Lancichinetti‐Fortunato‐Radicchi (LFR) networks. The results of our approach demonstrate improved efficiency and effectiveness when compared with the ten state‐of‐the‐art community detection algorithms. As a result of our findings, the proposed algorithm consistently outperforms the other network algorithms with different mixing parameters and degree distributions of LFR networks.