奇异向量空间双聚类算法

Singular Vector Space Biclustering Algorithm

本文针对0/1矩阵的双聚类问题提出一种奇异向量空间双聚类算法.通过SVD分解将0/1矩阵映射到左右奇异向量空间上,然后利用信息熵判断行聚类优先还是列聚类优先,最后根据判断结果递归进行行聚类或列聚类,直到满足停止条件.实验显示奇异向量空间双聚类算法可以分辨出完全无重叠的子矩阵,比较快速地得到硬的双簇.

In this paper,we propose a singular vector space biclustering algorithm for 0/1 binary clustering problems.At first,we compute the left and right singular vector matrices from the input 0/1 matrix,then the priority between the column-clustering and the row-clustering is determined based on information entropy.Finally,the column-clustering and the row-clustering are repeated iteratively until the stopping criterion is satisfied.Experimental results show that our algorithm can identify those completely non-overlapping sub-matrices and find the ＂hard＂ biclusters more efficiently.