The Neighborhood Role In The Linear Threshold Rank On Social Networks

Centrality and influence spread are two of the most studied concepts in social network analysis. Several centrality measures, most of them, based on topological criteria, have been proposed and studied. In recent years new centrality measures have been defined inspired by the two main influence spread models, namely, the Independent Cascade Model (IC-model) and the Linear Threshold Model (LT-model). The Linear Threshold Rank (LTR) is defined as the total number of influenced nodes when the initial activation set is formed by a node and its immediate neighbors. It has been shown that LTR allows to rank influential actors in a more distinguishable way than other measures like the PageRank, the Katz centrality, or the Independent Cascade Rank. In this paper we propose a generalized LTR measure that explore the sensitivity of the original LTR, with respect to the distance of the neighbors included in the initial activation set. We appraise the viability of the approach through different case studies. Our results show that by using neighbors at larger distance, we obtain rankings that distinguish better the influential actors. However, the best differentiating ranks correspond to medium distances. Our experiments also show that the rankings obtained for the different levels of neighborhood are not highly correlated, which validates the measure generalization.