The paper introduces a novel Modularized Nonnegative Matrix Factorization (M-NMF) model for network embedding, which aims to preserve both the microscopic structure (first- and second-order proximities of nodes) and the mesoscopic community structure. The M-NMF model integrates a NMF-based representation learning model and a modularity-based community detection model in a unified framework, leveraging the consensus relationship between node representations and community structures. The authors provide efficient updating rules for parameter inference and prove their correctness and convergence. Extensive experiments on various real-world networks demonstrate the superior performance of M-NMF over state-of-the-art methods in node clustering and classification tasks. The model's effectiveness is further validated through parameter analysis, showing its robustness to the number of communities.The paper introduces a novel Modularized Nonnegative Matrix Factorization (M-NMF) model for network embedding, which aims to preserve both the microscopic structure (first- and second-order proximities of nodes) and the mesoscopic community structure. The M-NMF model integrates a NMF-based representation learning model and a modularity-based community detection model in a unified framework, leveraging the consensus relationship between node representations and community structures. The authors provide efficient updating rules for parameter inference and prove their correctness and convergence. Extensive experiments on various real-world networks demonstrate the superior performance of M-NMF over state-of-the-art methods in node clustering and classification tasks. The model's effectiveness is further validated through parameter analysis, showing its robustness to the number of communities.