Probing Limits of Information Spread with Sequential Seeding

TL;DR

This paper introduces a sequential seeding method for information spread that outperforms traditional single-stage approaches, providing theoretical guarantees and empirical evidence of higher coverage with the same number of seeds.

Contribution

It proposes a coordinated randomized execution framework, proves the superiority of sequential seeding over single-stage methods, and demonstrates its effectiveness with simple degree-based selection.

Findings

Sequential seeding guarantees at least as much coverage as single-stage seeding.

Under certain conditions, sequential seeding achieves better coverage than single-stage with the same seeds.

Sequential seeding with degree-based selection surpasses greedy heuristic performance.

Abstract

We consider here information spread which propagates with certain probability from nodes just activated to their not yet activated neighbors. Diffusion cascades can be triggered by activation of even a small set of nodes. Such activation is commonly performed in a single stage. A novel approach based on sequential seeding is analyzed here resulting in three fundamental contributions. First, we propose a coordinated execution of randomized choices to enable precise comparison of different algorithms in general. We apply it here when the newly activated nodes at each stage of spreading attempt to activate their neighbors. Then, we present a formal proof that sequential seeding delivers at least as large coverage as the single stage seeding does. Moreover, we also show that, under modest assumptions, sequential seeding achieves coverage provably better than the single stage based approach using the same number of seeds and node ranking. Finally, we present experimental results showing how single stage and sequential approaches on directed and undirected graphs compare to the well-known greedy approach to provide the objective measure of the sequential seeding benefits. Surprisingly, applying sequential seeding to a simple degree-based selection leads to higher coverage than achieved by the computationally expensive greedy approach currently considered to be the best heuristic.