一种自适应优化相异性度量的基于MST的半监督聚类方法

It presents an MST-based semi-supervised clustering method of optimizing dissimilarity measure adaptively, when clustering an unlabelled data set which has the same or a similar distribution with a labelled sample in one hybrid attributes space. At first, we can obtain "regular cluster regions" by partitioning the labelled sample using one decision-tree method. Then optimize the dissimilarity measure of the hybrid attributes space adaptively based on the principia, "data points in the same clusters are close to each other". At last, the optimized dissimilarity measure is applied to an MST-based clustering method. This semi-supervised clustering method can often get a better clustering quality validated by some simulated experiments of several UCI data sets. In the end, it gives two research expectations in order to disinter and popularize this method.