Identifying Influential Spreaders in Social Networks Via Normalized Local Structure Attributes

Abstract

In social networks, influential spreaders are those nodes that can spread information to a large number of nodes. Identifying influential spreaders is a major challenge for applications, such as information diffusion acceleration, epidemic outbreak prevention, and effective e-commerce advertisement conduction. Existing methods consider the influence of a node’s neighbors on its spreading ability but rarely account for the topology of the neighboring nodes. Therefore, we propose a novel measure based on normalized local structure attributes, called normalized local centrality, which considers the topology of the local network around a node as well as the influence feedback of the node’s nearest neighbor nodes. First, we compute the influence of a node’s neighbors and the local clustering coefficient of them to identify nodes in cluster centers and those function as “bridge.” Then, a normalization function is designed to normalize the results to avoid adding new variable parameters. We perform experiments to identify influential spreaders in both real and computer-generated networks and compare the results on the basis of seven measures: degree, betweenness, closeness, k-shell, semi-local centrality, local structure centrality, and our proposed measure. In the susceptible–infected–recovered model, the node influence rankings obtained by our measure are most consistent with those of the benchmark, thus confirming that our method measures node influence more accurately than the other methods. Furthermore, the top-100 nodes ranked by our method lead to faster and wider spread than those ranked by the other six tested measures.

Keywords

• Social networks
• node spreading influence
• local centrality
• clustering coefficient