兼顾属性距离及关系强度的密度聚类算法

传统属性空间的密度聚类算法仅考虑对象属性取值相似度, 网络空间密度聚类算法仅关注对象间关系紧密度。针对两类算法的不足, 提出一种兼顾属性距离及关系强度的密度聚类算法。在构建兼顾属性距离及关系强度的网络之后, 完善了近邻对象及核心对象的概念, 并给出了相应的聚类策略。理论分析和实验结果表明, 由于综合考虑了属性、关系及关系强度信息, 算法规避了对象属性值分布对聚类过程的影响, 改善了聚类效果, 并能有效识别枢纽点和孤立点。

Density-based clustering method considering attribute distance and relationship strength

Traditional density-based clustering methods in attribute space just focus on the attribute distance between objects only, and the density-based clustering methods in network just use the relationship between objects. This paper proposed a density-based clustering method considering both attribute distance and relationship strength. After constructing the weighted network based on attribute distance and relationship strength, the algorithm refreshed the definition of near neighbor object and core object, and offered corresponding clustering policy. Theoretical analysis and test prove that the algorithm avoids the sad influence of the attribute value distribution, improves the clustering result, and distinguishes the hub and outlier objects which are useful in practice effectively.