基于引力模型的关键节点识别方法是常见的一类识别方法,但该类方法存在以下两方面不足,一是衡量节点“质量”时考虑的因素单一,二是所需的运行时间代价较高。通过优化这两方面不足,本文提出了一种改进的引力模型识别方法。首先,综合考...基于引力模型的关键节点识别方法是常见的一类识别方法,但该类方法存在以下两方面不足,一是衡量节点“质量”时考虑的因素单一,二是所需的运行时间代价较高。通过优化这两方面不足,本文提出了一种改进的引力模型识别方法。首先,综合考虑节点的度、H指数中心性和聚类系数以衡量节点“质量”,并利用信息熵对节点“质量”进行区分。然后,重新考虑节点的影响范围以减少所提方法的运行时间代价。最后,将本文提出的方法与六种基准方法在八个数据集上进行比较。实验结果表明,本文提出的方法在单调性、鲁棒性和准确性方面均表现出明显优势。与基于引力模型的识别方法相比,本文提出方法的运行时间可以降低18.97%~87.65%。The key node identification method based on gravity model is a common type of identification method, but this type of method has the following two shortcomings: the first is the single factor considered when measuring “mass” of the node;the second is the high cost of required running time. Therefore, this paper proposes an improved gravity model identification method by optimizing these two shortcomings. Initially, “mass” of the node is measured by comprehensively considering the degree, the H-index centrality and the clustering coefficient, and the information entropy is used to distinguish “mass” of the node. Then, the influence range of the node is being reconsidered to reduce the runtime cost of the proposed method. Finally, the method proposed in this paper is compared with six benchmark methods on eight datasets. Experimental results show that the proposed method has obvious advantages in monotonicity, robustness and accuracy. The running time of the proposed method in this paper can be reduced by 18.97% - 87.65% compared with the identification method based on the gravity model.展开更多
文摘基于引力模型的关键节点识别方法是常见的一类识别方法,但该类方法存在以下两方面不足,一是衡量节点“质量”时考虑的因素单一,二是所需的运行时间代价较高。通过优化这两方面不足,本文提出了一种改进的引力模型识别方法。首先,综合考虑节点的度、H指数中心性和聚类系数以衡量节点“质量”,并利用信息熵对节点“质量”进行区分。然后,重新考虑节点的影响范围以减少所提方法的运行时间代价。最后,将本文提出的方法与六种基准方法在八个数据集上进行比较。实验结果表明,本文提出的方法在单调性、鲁棒性和准确性方面均表现出明显优势。与基于引力模型的识别方法相比,本文提出方法的运行时间可以降低18.97%~87.65%。The key node identification method based on gravity model is a common type of identification method, but this type of method has the following two shortcomings: the first is the single factor considered when measuring “mass” of the node;the second is the high cost of required running time. Therefore, this paper proposes an improved gravity model identification method by optimizing these two shortcomings. Initially, “mass” of the node is measured by comprehensively considering the degree, the H-index centrality and the clustering coefficient, and the information entropy is used to distinguish “mass” of the node. Then, the influence range of the node is being reconsidered to reduce the runtime cost of the proposed method. Finally, the method proposed in this paper is compared with six benchmark methods on eight datasets. Experimental results show that the proposed method has obvious advantages in monotonicity, robustness and accuracy. The running time of the proposed method in this paper can be reduced by 18.97% - 87.65% compared with the identification method based on the gravity model.
