Automatic Kernel Parameter Tuning of KSC for Video Category Classification

Here we focus on website video-tag category clustering using kernel spectral clustering (KSC) [1], which has not been reported so far. KSC is a nonlinear clustering method used for the spectral clustering of data in a kernel feature space. Unfortunately KSC is intrinsically affected by the selection of kernel parameters, and this hinders complete automatic learning. This study proposes a unified learning method which enables automatic KSC optimal kernel parameter tuning based on a graph Laplacian representation of the clustering. This method is well suited to applications that require speedy and fully automatic analysis, since it can automatically find the optimal arameter, and generate the optimally connected components of the graph Laplacian. We demonstrate the application of our method to the above problem and show its excellent performance through computer experiments.