Using Link Clustering to Detect Influential Spreaders

Abstract

Spreading processes are increasingly analysed in the context of complex networks, for example in epidemics research, information dissemination or rumors. In these contexts, the effect of structural properties that facilitate or decelerate spreading processes are of high interest, since this allows insights into the extent to which those processes are controllable and predictable. In social networks, actors usually participate in different densely connected social groups that emerge from various social contexts, such as workplace, interests, etc. In this paper, it is examined if the number of groups an actor connects to can be used as a predictor for its capability to spread information effectively. The social contexts (i.e. groups) a node participates in are determined by the Link Communities approach by Ahn et al. (2010). The results are contrasted to previous findings of structural node properties based on the k-shell index of nodes (Kitsak et al. 2010) by applying both methods on artificially generated and real-world networks. They show that the approach is comparable to existing ones using structural node properties as a predictor, yet no clear evidence is found indicating that one or the other approach leads to better predictions in all investigated networks.