一种新的平衡化谱聚类方法

A New Balanced Spectral Clustering Method

针对传统谱聚类算法在非平衡数据集上聚类效果不理想的问题,提出了一种平衡化谱聚类算法,该算法在传统谱聚类目标函数的基础上加入了对聚类隶属度矩阵的近似正交约束,从而得到新的聚类目标函数.实验结果表明,新算法可以缓解传统谱聚类产生的均匀效应,提升了在非平衡数据集上的聚类纯度.

Aiming at the problem that the traditional spectral clustering algorithm has poor clustering effect on unbalanced data sets, a balanced spectral clustering algorithm is proposed.The algorithm adds an approximate orthogonal constraint to the clustering membership matrix on the basis of the traditional spectral clustering objective function to obtain a new clustering objective function.The experimental results show that the new algorithm can alleviate the uniform effect of traditional spectral clustering and improve the clustering purity on unbalanced datasets.