基于属性约简集评价节点重要性研究

社会网络成员的重要性确定通常依赖结构属性对网络节点的评价。首先定义了网络中节点排序可区分以及属性约简集的概念,并在此基础上量化了属性聚类的阈值,从而确定了类别的数量。设计了网络节点重要性的属性约简集评价算法。通过与度、介数、全属性评价在人工网络、海豚网上的实现,证明了属性约简集评价节点排序的可行性。通过属性约简集在海豚网、9·11恐怖分子合作网上的节点评价值、网络鲁棒性以及节点可区分性等方面的应用对比分析,发现属性约简集评价节点重要性既兼顾了网络结构的完整性,又避免了单一属性评价的片面性和多个属性之间的属性冗余性,提高了节点评价结果的准确性,降低了算法复杂度。

Node Importance Rank by Attribute Reduction Set Evaluation

The nodes importance evaluation usually depends on the structure properties. In this paper, two parameters, discriminable rank and attribute reduction evaluation on nodes, are defined. A new algorithm by attributes set reduction evaluation is designed to rank nodes importance. Comparing with degree-based, betweenness-based, total attribute-based evaluations, the advantages of attribute set reduction evaluation algorithm, such as the higher effectiveness and more accuracy are displayed in application on Dolphin network and 9·11 Terrorist Attack network. Moreover, the attribute set reduction evaluation algorithm displays more superiority not only on keeping global structure properties but also on avoiding attributes redundancy.