一种基于MST的自适应优化相异性度量的半监督聚类方法

针对混合属性空间中具有同一(或相近)分布特性的带类别标记的小样本集和无类别标记的大样本数据集,提出了一种基于MST的自适应优化相异性度量的半监督聚类方法。该方法首先采用决策树方法来获取小样本集的"规则聚类区域",然后根据"同一聚类的数据点更为接近"的原则自适应优化建构在该混合属性空间中的相异性度量,最后将优化后的相异性度量应用于基于MST的聚类算法中,以获得更为有效的聚类结果。仿真实验结果表明,该方法对有些数据集是有改进效果的。为进一步推广并在实际中发掘出该方法的应用价值,本文在最后给出了一个较有价值的研究展望。

A Semi-Supervised Clustering Method of Adaptively Optimizing the Dissimilarity Based on MST

This paper presents an MST-based semi-supervised clustering method of adaptively optimizing dissimilarity,when clustering an unlabeled data set which has the same or a similar distribution with a labeled sample in one hybrid attributes space.First,we can obtain "regular cluster regions" by using a decision-tree method,and then adaptively optimize the dissimilarity of the hybrid attributes space based on the principia,"data points in the same clusters should have more similarity than those in other clusters".Finally,the optimized dissimilarity is applied to an MST-based clustering method.From some simulated experiments of several UCI data sets,we know that this kind of semi-supervised clustering method can often get better clustering quality.In the end,it gives a research expectation to disinter and popularize this method.