In the process of graph clustering, the quality requirements for the structure of data graph are very strict, which will directly affect the final clustering results. Enhancing data graph is the key step to improve the performance of graph clustering. In this paper, we propose a self-adaptive clustering method to obtain a dynamic fine-tuned sparse graph by learning multiple static original graph with different sparsity degrees. By imposing a constrainted rank on the corresponding Laplacian matrix, the method utilizes the eigenvectors of the Laplacian matrix to create a new and simple data sparse matrix to have exactly k connected components, so that the method can quickly and directly learn the clustering results. The experimental results on synthetic and multiple public datasets verify that the proposed method is meaningful and beneficial to discover the real cluster distribution of datasets.

在图聚类的过程中，对数据图结构的质量要求非常严格，这将直接影响最终的聚类结果。增强数据图是提高图聚类性能的关键步骤。本文提出了一种自适应聚类方法，通过学习具有不同稀疏度的多个静态原始图来获得动态微调的稀疏图。通过在相应的拉普拉斯矩阵上施加约束秩，该方法利用拉普拉斯矩阵的特征向量来创建新的简单数据稀疏矩阵，使其具有正好为k连接的组件，以便该方法可以快速直接地了解聚类结果。综合和多个公共数据集的实验结果证明，该方法是有意义的，有利于发现数据集的真实聚类分布。