【摘要】 复杂网络中的重要节点发现在现实生活中有着广泛的应用价值。传统重要节点发现方法可分为局部发现和全局发现两类算法,全局发现算法中最具代表性的是特征向量中心性算法(Eigenvector Centrality, EC),EC算法将所有节点归为一个社区并利用邻居节点重要性反馈计算节点的影响力大小,具有较高的计算效率和识别精度。但是,EC算法忽略了网络的拓扑结构,未考虑到真实网络中节点所在社区的结构特征。为此,本文提出一种基于网络拓扑结构的可达中心性算法(Accessibility Centrality, AC),首先利用邻接矩阵作为反馈路径,在反馈过程中计算不同路径下的节点整体影响力。同时,利用影响力传递过程中的噪音干扰特性,修正每一路径长度下节点整体影响力大小,最后利用修正结果得到AC值。为评估AC算法,本文利用两种传染病模型模拟节点影响力在四组真实网络中的传播过程,并引入其他四种算法进行对比验证。实验结果表明,与其他算法相比,AC算法可以更准确、有效地识别出有具有影响力的重要节点。更多还原

【Abstract】 Identifying influential nodes in complex networks has been widely applied in reality. Conventional methods consist of local metrics and global metrics, and Eigenvector Centrality(EC)algorithm is the most representative method in global metrics. EC algorithm regards all nodes in network as one community and calculates the influence of nodes by considering the node influence of their neighbors. However, EC algorithm ignores the topology structure of community networks in reality, where nodes’ neighbors may come from different communities with different clustering coefficient. To address this issue, this paper proposes an Accessibility Centrality(AC)model based on the network topology. Firstly, the adjacency matrix is considered as the feedback path, and nodeinfluence is computed through different pathswith different length. Then, node influence from every step is revised, due to existing noise in the paths. At last, every nodes’ influence(AC value)is calculated via the revised result in the previous procedure. To evaluate the performance of our method, Susceptible-Infected-Recovered(SIR)model and Susceptible-Infected(SI)model are employed to simulate the process of influence propagation on four real networks. Comparing with Eigenvector Centrality, Betweenness Centrality, Degree Centrality and LeaderRank algorithms, the results show that our method obtains the highest accuracy.更多还原