基于二部图矩阵的一致性多视图聚类

CONSISTENT MULTI-VIEW CLUSTERING BASED ON BIPARTITE GRAPH MATRIX

为了充分利用一致性信息,提升聚类性能与自适应能力,提出一种基于二部图矩阵的一致性多视图聚类方法。通过少量代表不同视图的统一锚定点表示一致性信息,将它们融合生成统一的图矩阵。另外以一种相互加强的方式确定连续锚点,从而自动确定每个二部图的权重。进一步通过交替优化逐步求解变量优化问题。在合成数据集和真实数据集上的实验结果证明了该方法在聚类精度与自适应能力上的优越性。

In order to make full use of consistency information and improve clustering performance and adaptive ability,a consistent multi-view clustering method based on bipartite graph matrix is proposed.A small number of unified anchors representing different views were used to represent the consistency information.Those information were fused to form a unified graph matrix.In addition,the weight of each bipartite graph was automatically determined by determining continuous anchors in a mutually reinforcing way.Further,the variable optimization problem was solved step by step by alternating optimization.The experimental results on synthetic data sets and real data sets prove the superiority of the proposed method in clustering accuracy and adaptive ability.