Influence Maximization via Graph Neural Bandits

TL;DR

The paper addresses Influence Maximization under unknown diffusion topologies by casting it as a contextual bandit problem and introducing IM-GNB, which builds per-arm exploitation and exploration graphs from learned user predictors and refines seed-reward estimates with GNNs. By integrating gradient-informed exploration, dual graphs, and GCN-based reward estimation, IM-GNB achieves improved diffusion spread in multi-round campaigns on real datasets without assuming diffusion models. Key contributions include the dual-graph construction via $h_u^{(1)}$ and $h_u^{(2)}$, GNN-based exploitation and exploration nets $f^{(1)}$ and $f^{(2)}$, and a scalable complexity framework with clustering and pooling. The approach offers a practical, model-agnostic solution for influencer marketing and information campaigns in uncertain networks, showing robust performance across contexts and seed-budget settings.

Abstract

We consider a ubiquitous scenario in the study of Influence Maximization (IM), in which there is limited knowledge about the topology of the diffusion network. We set the IM problem in a multi-round diffusion campaign, aiming to maximize the number of distinct users that are influenced. Leveraging the capability of bandit algorithms to effectively balance the objectives of exploration and exploitation, as well as the expressivity of neural networks, our study explores the application of neural bandit algorithms to the IM problem. We propose the framework IM-GNB (Influence Maximization with Graph Neural Bandits), where we provide an estimate of the users' probabilities of being influenced by influencers (also known as diffusion seeds). This initial estimate forms the basis for constructing both an exploitation graph and an exploration one. Subsequently, IM-GNB handles the exploration-exploitation tradeoff, by selecting seed nodes in real-time using Graph Convolutional Networks (GCN), in which the pre-estimated graphs are employed to refine the influencers' estimated rewards in each contextual setting. Through extensive experiments on two large real-world datasets, we demonstrate the effectiveness of IM-GNB compared with other baseline methods, significantly improving the spread outcome of such diffusion campaigns, when the underlying network is unknown.

Figures (4)

• Figure 1: The framework of IM-GNB. For each arm, we initially take the arm feature vector and the current context vector $(\bm{k}_i,\bm{C}_t)$ as inputs to estimate the diffusion probability for each user-arm pair with $h_u^{(1)}$. Subsequently, we assess the potential gain on the diffusion probability with the past gradient of $h_u^{(1)}$, yielding both exploitation and exploration graphs. With the pre-estimated graphs, we refine the estimate of the diffusion probability for each user-arm pair with $f^{(1)}$ and $f^{(2)}$. The aggregate reward of the arm across all users is derived from the sum of all the refined individual diffusion probabilities. The potential gain is measured similarly. Finally, we select the arm with the highest sum of estimated reward and its potential gain.
• Figure 2: Comparison of IM-GNB with baselines on the Twitter dataset.
• Figure 3: Comparison of IM-GNB with baselines on the Weibo dataset.
• Figure 4: Analysis on the number of clustering groups.