搜索

x
中国物理学会期刊

基于传播方向冗余抑制的复杂网络多影响力节点识别方法

Identifying multiple influential nodes in complex networks based on propagation-direction redundancy suppression

PDF
导出引用
  • 在舆情导控、病毒营销及疫情阻断等复杂网络应用中,多影响力节点识别需要兼顾种子节点的传播能力和种子集合内部的覆盖重合。本文从传播方向的异质性出发,提出一种基于传播方向冗余抑制(propagation-direction redundancy suppression, PDRS)的多影响力节点识别算法。 PDRS利用传播方向权重分布的熵定义有效方向指数,以衡量支撑节点传播潜能的有效方向数。每当选出一个种子节点后,算法根据相邻候选节点与该种子的传播方向重合程度动态削减候选得分,使后续种子分布到尚未充分覆盖的网络区域。本文在 9 个真实无向网络和 9 个真实有向网络上采用易感、感染和恢复模型(susceptible-infected-recovered, SIR)开展实验。 PDRS 取得较高的平均最终感染比例,并减少种子节点之间的局部覆盖重合。

     

    Many applications involving complex networks, including public opinion guidance and control, viral marketing, and epidemic containment, require the identification of multiple influential nodes. An effective seed set should combine strong spreading ability with limited overlap in coverage. Methods based mainly on individual-node influence may select several seeds from the same densely connected region; this concentration limits the additional coverage of later selections. To address this problem, we propose a propagation-direction redundancy suppression (PDRS) algorithm that accounts for heterogeneity among a node's propagation directions. PDRS assigns each direction a weight based on the connectivity of the neighboring node and the redundancy indicated by common neighbors. The entropy of the normalized weight distribution defines an effective direction index that quantifies the effective number of low-redundancy directions supporting propagation. After selecting a seed, PDRS reduces each adjacent candidate's score according to the share of its directional weight assigned to directions that overlap the seed's local coverage. The candidate's local clustering coefficient adjusts the reduction. Subsequent seeds are therefore selected from less-covered regions. For directed networks, two-hop reachability quantifies subsequent spreading capacity, while local reciprocity adjusts the score reduction. We compare PDRS with 13 methods under the susceptible-infected-recovered (SIR) model on nine real-world undirected networks and nine real-world directed networks. Under these SIR settings, PDRS achieves the highest overall mean final infection proportion among the methods compared and reduces local coverage overlap among selected seeds. Ablation results show that the effective direction index changes the initial ranking, whereas the dynamic score update contributes more to the observed improvements in spreading performance and seed dispersion. PDRS therefore balances the spreading ability of individual seeds with the suppression of redundant coverage within the seed set.

     

    目录

    /

    返回文章
    返回