Accelerating Simulation-Based Influence Maximization via Bayesian Optimization

Zonghan Zhang, Zhiqian Chen

Primary Area: learning on graphs and other geometries & topologies

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: influence maximization, Bayesian optimization, simulation

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Use Bayesian optimization to build a relationship between the seed set and its influence spread evaluated by simulations.

Abstract: Influence Maximization (IM) has garnered significant attention due to its broad applicability in areas such as viral marketing, social network recommendations, and disease containment. The primary goal of IM is to identify an optimal seed set that maximizes influence spread. Existing methodologies for IM are largely categorized into proxy-based and simulation-based approaches, each with its own limitations. Proxy-based methods often fail to capture complex seed interactions and are model-specific, while simulation-based techniques are computationally expensive for large-scale graphs. Additionally, current research lacks a comprehensive model to understand the relationship between seed set configurations and their resulting influence spreads. To address these challenges, we present a Bayesian Optimization Influence Maximization (BOIM) framework that employs Bayesian optimization to minimize the number of required simulations. Our approach utilizes a Gaussian Process (GP) as the surrogate function to model the complex interplay between seed sets and influence spreads. In GP, we also introduce a specialized kernel for graph-level Bayesian optimization and implement stratified sampling to ensure uniform instance distribution. Our methodology offers a computationally efficient yet accurate alternative to traditional IM approaches, effectively bridging the gap between computational efficiency and approximation fidelity. Extensive experimentation has demonstrated that our approach has effectiveness and efficiency that surpasses standard simulation methods.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 4668