Online Influence Maximization via an Explore-exploit Ensemble Approach
Abhinav Choudhury, Prakash Duraisamy, James Van Haneghan, Fang Jiang, Varun Dutt · 2024
Online Influence Maximization (OIM) seeks to maximize influence spread by identifying the optimal individuals to expose information to, despite lacking prior knowledge of influence probabilities. Traditional OIM algorithms often receive limited feedback, as they only partially explore networks. To address this, we propose an explore-exploit ensemble approach based on the EXP4 algorithm, introducing two variants: EXP4-OIM and EXP4-OIM-Greedy. Tested on the NetHEPT (15,229 nodes, 62,752 edges) and PhysicianSN (181 nodes, 19,026 edges) datasets, both EXP4OIM and EXP4-OIM-Greedy outperformed existing benchmarks in influence spread. This research demonstrates the effectiveness of our ensemble approach in optimizing information dissemination in social networks.