Key Node Identification Method Integrating Information Transmission Probability and Path Diversity in Complex Network

Abstract

Abstract Previous key node identification approaches assume that the transmission of information on a path always ends positively, which is not necessarily true. In this paper, we propose a new centrality index called Information Rank (IR for short) that associates each path with a score specifying the probability that such path successfully conveys a message. The IR method generates all the shortest paths of any arbitrary length coming out from a node $u$ and defines the centrality of u as the sum of the scores of all the shortest paths exiting $u$. The IR algorithm is more robust than other centrality indexes based on shortest paths because it uses alternative paths in its computation, and it is computationally efficient because it relies on a Beadth First Search-BFS to generate all shortest paths. We validated the IR algorithm on nine real networks and compared its ability to identify super-spreaders (i.e. nodes capable of spreading an infection in a real network better than others) with five popular centrality indices such as Degree, Betweenness, K-Shell, DynamicRank and PageRank. Experimental results highlight the clear superiority of IR over all considered competitors.