<p>Community detection is a fundamental task for understanding the structure and functionality of complex networks. Most existing methods rely primarily on direct topological connections between nodes, often neglecting both intrinsic node attributes and latent structural information among indirectly connected nodes. This limitation reduces the accuracy and robustness of the resulting community partitions. To address this issue, we propose a non-negative matrix factorization method based on attribute and latent structure information (ALSINMF) for community detection. Specifically, ALSINMF introduces Katz centrality to capture potential relationships among non-adjacent nodes and applies sparsity constraints to the community attribute matrix, enhancing inter-community distinctiveness. Experimental results demonstrate that ALSINMF consistently outperforms several recent baselines across five real-world data sets, achieving notable improvements in accuracy, Jaccard similarity, and nmi. Moreover, the method maintains stable superiority on both small-scale WebKB data sets and the larger, attribute-rich Citeseer data set, confirming its effectiveness and robustness.</p>