Abstract
We study the problem of identifying top-K influencers when we have only local knowledge of the network structure. More specifically, the selection of top-K influencers is performed sequentially over a number of rounds. We propose an efficient algorithm called strength network similarity-based upper confidence bound (SNS_UCB1) for the identification of top-K influencers based on upper confidence bound (UCB1) from the multi-armed bandit's framework. Considering feedback in online decision-making, we rely on edge (arm) strength on falling within a large number of other edges and how edge members are similar to each other and can thus convince other users to adopt the promoted behaviours. Thus, this feedback is considered as a reward score at each pull of the arm of how likely this selection is to contribute to the increase in the cumulative reward. We evaluate the proposed algorithm under the independent cascade (IC) model on four large-scale datasets that differ in size and density. We compare our algorithm to a centrality measure-based UCB1 and several well-known state-of-the-art approaches, demonstrating its superior performance in terms of influence spread achieved with the less required time and storage space.