链接文档中基于子空间分解的高效谱聚类算法

Efficient spectral clustering algorithm based on subspace decomposition in linked documents

摘要:

提出了一种基于子空间分解的高效谱聚类算法。首先，基于共识信息和特定域信息的矩阵分解将链接文档划分为3个子空间，然后对子空间添加正则化项建模共识信息和特定域信息对聚类的不同影响，并采用交替优化方法实现谱聚类。考虑到谱聚类的复杂性，提出了一种带曲线搜索的梯度下降法加速求解过程。3个真实数据集上的实验结果表明，所提算法在聚类质量和效率方面始终明显优于目前典型的基线算法，且对输入参数不敏感。

Abstract:

An efficient spectral clustering algorithm based on subspace decomposition is proposed. Firstly, based on the matrix decomposition of consensus information and domain specific information, the linked documents are divided into three subspaces, the different effects of consensus information and domain specific information on clustering are modeled by adding regularization items to the subspaces, and the alternative optimization method is utilized to achieve spectral clustering. In addition, considering the complexity of spectral clustering, a gradient descent method with curvilinear search is proposed to accelerate the solution process. Experimental results on three real datasets show that the proposed algorithm is superior to the current typical baseline algorithm in terms of clustering quality and efficiency, and is insensitive to input parameters.