Affinity Propagation Clustering Based on Variable-Similarity Measure
可变相似性度量的近邻传播聚类

Affinity Propagation (AP) clustering is not fit to deal with multi-scale data cluster as well as the arbitrary shape cluster issue. Therefore, an improved affinity propagation clustering algorithm AP-VSM (Affinity Propagation based on Variable-Similarity Measure) is proposed embarking from the token of data distribution characters. First, a kind of variable-similarity measure method is devised according of characters of global and local data distribution, which has the ability of describing the characters of data clustering effectively. Then AP-VSM clustering algorithm is proposed base on the frame of traditional AP algorithm, and this method has extended data processing capacity compared with traditional AP. The simulation results show that the new method is outperforming traditional AP algorithm.