Semi-supervised Spectral Clustering Algorithm Based on Bayesian Decision ⋆

Recently, semi-supervised spectral clustering algorithms have been developing rapidly, which are proposed to improve the clustering performance. In this paper, we first review the current existing spectral clustering algorithms in an unified-framework and give a straightforward explanation about the spectral clustering algorithm. Then, we present a semi-supervised method to improve the clustering results, whose basic idea is to adjust the similarity matrix based on Bayesian information, and fix the class labels on the reference of the pairwise constraints at last. Experimental results on the UCI datasets demonstrate the advantages of the method. We conclude that the similarity adjustment method built on Bayesian decision is meaningful, the boundary of the clusters is partitioned more correctly.