Sequential Seeding in Complex Networks: Trading Speed for Coverage.

Diffusion in complex networks studied in the paper is a process of spreading information, ideas or substances with replication between nodes through edges from the regions of high to low concentration. Diffusion cascades are triggered by the activation of a small set of initial nodes - seeds - and later supported by the natural process. In this work, several novel approaches related to extension of the commonly used seeding strategies into a sequence of stages are proposed. Sequential seeding method is compared with a single stage approach using both real and artificial complex networks and applying various dynamic node ranking methods and diffusion parameters. The experimental results show that sequential seeding strategies deliver better results than single stage seeding in most cases. These strategies avoid seeding nodes that are activated through the natural diffusion process at the preceding stages of sequential seeding. The gain arises when a saved seed is allocated to a node difficult to reach via diffusion process. The performance is further improved by identifying non-activated nodes with high potential of activizing their not yet active neighbors. It is boosted by adaptive recomputing of network measures on the subgraph with non-activated nodes only. The higher length of seeding sequences tends to activate more nodes but extends the duration of diffusion process. The results demonstrate the trade-off between the coverage and speed of diffusion that can be balanced according to the importance assigned to coverage and time in specific applications.