Research map / Graph learning / Graph algorithms and clustering
Graph Bayesian Optimization for Multiplex Influence Maximization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Formulates multiplex influence maximization over multiple associated information items and proposes GBIM, a graph Bayesian optimization framework with a global kernelized attention message-passing surrogate and Bayesian linear regression.
Problem
Influence maximization ignores simultaneous, interacting diffusion of multiple information items; selecting user-item seed pairs under a budget to maximize multiplex influence has huge combinatorial and computational complexity.
Approach
Combine a linear-complexity global kernelized attention message-passing network (positive random features) as a learned basis with Bayesian linear regression and an expected-improvement acquisition with explore-exploit sampling, to optimize user-item seed sets under a multiplex diffusion model with an association mechanism.
Key results
- GBIM consistently outperforms IMM, CELF++, DeepIM, BO and MaxDegree across seed set sizes and datasets on Multi-LT and Multi-IC.
- On LastFM, GBIM attains over 40% higher multiplex influence spread than the best baseline.
- Scalability: near-linear runtime in number of users and stable runtime as number of items grows; IMM and CELF++ exceed time/memory limits on several settings.
Surprising findings
- Standard Gaussian-process BO performs mediocre relative to the proposed method, and DeepIM falls short across all datasets.
Limitations
- Item relations are limited to a single association mechanism; complementary and competitive relations are not learned from data.
- Results rely on simulated diffusion models (Multi-LT, Multi-IC) rather than observed cascades.
Experimental setup
Datasets: Ciao Epinions Delicious LastFM Synthetic Erdos-Renyi graph (30,000 nodes)
Benchmarks: Ciao Epinions Delicious LastFM Synthetic
Metrics: expected multiplex influence (average over 100 simulations) runtime
Baselines: IMM CELF++ DeepIM BO (Gaussian process) MaxDegree
Builds on: Bayesian optimization Performer (positive random features) DeepIM IMM CELF++
Closest papers by meaning
- REM: A Scalable Reinforced Multi-Expert Framework for Multiplex Influence Maximization (AAAI 2025 · similarity 0.91)
- DiffIM: Differentiable Influence Minimization with Surrogate Modeling and Continuous Relaxation (AAAI 2025 · similarity 0.88)
- Unveiling Environmental Sensitivity of Individual Gains in Influence Maximization (NeurIPS 2025 · similarity 0.87)
- Motif-oriented influence maximization for viral marketing in large-scale social networks (NeurIPS 2024 · similarity 0.86)
- Hyperparametric Robust and Dynamic Influence Maximization (AAAI 2025 · similarity 0.85)
- DeepSN: A Sheaf Neural Framework for Influence Maximization (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.