A parameterizable influence spread-based centrality measure for influential users detection in social networks

• Published in: Knowledge-based systems 2022-12, Vol.257, p.109922, Article 109922
• Main Authors: Riquelme, Fabián, Vera, José-Antonio
• Format: Article
• Language: English
• Subjects: Centrality; Influence spread; Social influence; Social network
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2022.109922

In social network analysis, centrality refers to the relevance of actors or nodes within a social network represented as a graph. Traditional centrality measures are based on topological aspects of the network or the information flow circulating through it. Since 2010, new centrality measures have been proposed based on the influence spread capacity of the actors. One of the purposes of centrality measures is to obtain well-differentiated rankings among the actors. In order to achieve the above, recent experiments suggest that the actors must rely on some neighbors to increase their initial influence spread capacity. This article presents the General Influence Spread Rank (GISR), a generalized and parameterizable centrality measure based on two well-known influence spread models: the Independent Cascade model and the Linear Threshold Model. This measure allows adjusting the depth levels, probability, and directionality of the initial activation of neighbors, to study how these parameters intervene in the actors’ centrality. The experiments validate the importance of the depth level of neighborhoods and reveal that other connectivity aspects are also relevant to the centrality problem. Further, the depth level seems to be the most relevant variable in the influence spread for denser networks. In contrast, for sparser networks, the probability of selecting neighbors seems even more significant than the depth level. •We propose a new centrality measure based on influence spread models.•This measure accepts different parameters regarding the neighborhoods activation.•We apply two experiments with 257 executions using different parameter settings.•For well-connected networks, the depth level of neighborhoods is critical.•For loosely connected networks, the initial activation probability of the neighbors is critical.