Abstract: To address the inconsistency between attribute completion and clustering objectives in two-stage methods and the low efficiency of deep learning approaches for attribute-missing graph clustering, this paper proposes a stacked joint optimization-based algorithm. Firstly, a Matrix Factorization-based Attribute-Missing Graph Clustering Model (AMGC-MF) is constructed, which introduces an enhanced adjacency matrix to represent global node relations and employs graph regularization for attribute completion. Joint non-negative matrix factorization is then applied to the completed attribute matrix and the enhanced adjacency matrix to learn the cluster membership. Building on this, an Attribute-Missing Graph Clustering Model based on Stacked Joint Optimization (AMGC-SJO) is designed to dynamically update both matrices during iteration, enabling co-optimization of attribute completion and clustering. Experimental results show that AMGC-SJO outperforms AMGC-MF across multiple clustering metrics, with an average improvement of at least 0.68 percentage points on five datasets. Compared to the deep learning method AMGC, AMGC-SJO achieves comparable or better accuracy while reducing runtime by at least 89.57%. Moreover, under high missing rates (90%), AMGC-SJO demonstrates strong robustness with performance fluctuations within 4.35 percentage points, whereas AMGC-MF and AMGC decline by over 15 and 37 percentage points, respectively. The proposed AMGC-SJO offers an accurate, efficient, and robust solution for attribute-missing graph clustering, with significant theoretical and practical implications for large-scale real-world network analysis.

Key words: Keywords: attribute-missing graph, clustering, non-negative matrix factorization, graph regularization

摘要： 摘 要: 针对属性缺失图聚类中两阶段方法的目标不一致性及深度学习方法效率低的问题，本文提出一种基于堆叠式联合优化的属性缺失图聚类算法。首先，构建基于矩阵分解的属性缺失图聚类模型（AMGC-MF），通过增强邻接矩阵刻画节点间全局连接关系，并借助图正则化实现属性补全，进而对补全后的属性矩阵与增强邻接矩阵进行联合非负矩阵分解，以学习节点聚类隶属度。在此基础上，进一步设计堆叠式联合优化模型（AMGC-SJO），在迭代优化中动态更新增强邻接矩阵与属性矩阵，实现属性补全与聚类任务的协同优化。实验结果表明，AMGC-SJO在多项聚类评估指标上显著优于AMGC-MF，在五个数据集上关键指标平均提升不低于0.68个百分点；与深度学习方法AMGC相比，在保持相当聚类精度的同时，运行时间降低至少89.57%。此外，在高缺失率（90%）场景下，AMGC-SJO表现出优异鲁棒性，性能波动不超过4.35个百分点，而AMGC-MF与AMGC分别下降超过15与37个百分点。本文所提方法为属性缺失图聚类提供了一种高精度、高效率、强鲁棒的解决方案，对大规模真实网络数据处理具有理论与应用价值。

关键词: 关键词: 属性缺失图, 聚类, 非负矩阵分解, 图正则