基于属性——关系综合相似度的聚类算法

Clustering algorithm based on attribute-relationship integrated similarity

为弥补属性空间聚类方法只关注对象属性信息以及结构聚类方法只关注对象间关系信息的不足,提出一种基于属性—关系综合相似度的聚类算法。在构建基于属性距离的有权网络后,算法给出对象间综合相似度以及类间综合相似度的计算方法,并设计相应策略自底向上实现聚类。与属性空间聚类和结构聚类方法相比,该算法由于兼顾了属性和关系信息而具有更高的聚类精度。与经典的基于属性—关系相似度的算法HyPursuit、M-S等相比,算法由于精简了综合相似度的计算而具有更高的效率。

In order to improve the attribute space clustering methods that focus on the attribute information only, and improve the structure clustering methods that focus on relationship only ,this paper proposed a clustering algorithm based on attribute-re- lationship integrated similarity. After setting up the weighted network based on attribute distance, the algorithm showed how to calculate the integrated similarity between objects and between clusters, and provided appropriate strategy to aggregating clusters from bottom up. In comparison with attribute space clustering methods and structure clustering methods, the algorithm has higher accuracy for considering more information. In comparison with classic method based on attribute-relationship similarity like HvPursuit or M-S, the algorithm has better efficiency for simplifving the computing process of integrated similaritv.