Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Task Combinatorial Bandits for Budget Allocation

Lin Ge, Yang Xu, Jianing Chu, David Cramer, Fuhong Li, Kelly Paulson, Rui Song

TL;DR

The paper addresses budget allocation across many advertising campaigns under uncertain returns by framing the problem as a multi-task combinatorial bandit. It introduces a Bayesian hierarchical CMAB that shares information across campaigns through a global function $g(oldsymbol{x}_{m,k},a)$ and arm-specific random effects, with Thompson sampling guiding online decisions. Offline simulations and real campaign data show faster convergence and higher cumulative rewards, including substantial gains in clicks, while an online A/B test demonstrates a 12.7% reduction in cost-per-click. The work suggests promising directions for incorporating contextual seasonality and cross-campaign competition to achieve globally optimal budget utilization.

Abstract

Today's top advertisers typically manage hundreds of campaigns simultaneously and consistently launch new ones throughout the year. A crucial challenge for marketing managers is determining the optimal allocation of limited budgets across various ad lines in each campaign to maximize cumulative returns, especially given the huge uncertainty in return outcomes. In this paper, we propose to formulate budget allocation as a multi-task combinatorial bandit problem and introduce a novel online budget allocation system. The proposed system: i) integrates a Bayesian hierarchical model to intelligently utilize the metadata of campaigns and ad lines and budget size, ensuring efficient information sharing; ii) provides the flexibility to incorporate diverse modeling techniques such as Linear Regression, Gaussian Processes, and Neural Networks, catering to diverse environmental complexities; and iii) employs the Thompson sampling (TS) technique to strike a balance between exploration and exploitation. Through offline evaluation and online experiments, our system demonstrates robustness and adaptability, effectively maximizing the overall cumulative returns. A Python implementation of the proposed procedure is available at https://anonymous.4open.science/r/MCMAB.

Multi-Task Combinatorial Bandits for Budget Allocation

TL;DR

The paper addresses budget allocation across many advertising campaigns under uncertain returns by framing the problem as a multi-task combinatorial bandit. It introduces a Bayesian hierarchical CMAB that shares information across campaigns through a global function and arm-specific random effects, with Thompson sampling guiding online decisions. Offline simulations and real campaign data show faster convergence and higher cumulative rewards, including substantial gains in clicks, while an online A/B test demonstrates a 12.7% reduction in cost-per-click. The work suggests promising directions for incorporating contextual seasonality and cross-campaign competition to achieve globally optimal budget utilization.

Abstract

Today's top advertisers typically manage hundreds of campaigns simultaneously and consistently launch new ones throughout the year. A crucial challenge for marketing managers is determining the optimal allocation of limited budgets across various ad lines in each campaign to maximize cumulative returns, especially given the huge uncertainty in return outcomes. In this paper, we propose to formulate budget allocation as a multi-task combinatorial bandit problem and introduce a novel online budget allocation system. The proposed system: i) integrates a Bayesian hierarchical model to intelligently utilize the metadata of campaigns and ad lines and budget size, ensuring efficient information sharing; ii) provides the flexibility to incorporate diverse modeling techniques such as Linear Regression, Gaussian Processes, and Neural Networks, catering to diverse environmental complexities; and iii) employs the Thompson sampling (TS) technique to strike a balance between exploration and exploitation. Through offline evaluation and online experiments, our system demonstrates robustness and adaptability, effectively maximizing the overall cumulative returns. A Python implementation of the proposed procedure is available at https://anonymous.4open.science/r/MCMAB.
Paper Structure (14 sections, 5 equations, 3 figures, 1 table, 1 algorithm)

This paper contains 14 sections, 5 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Budget Allocation
  5. Combinatorial Multi-Armed Bandits
  6. Methodology
  7. Problem Formulation
  8. Multi-task Bayesian Hierarchical CMAB Framework
  9. Learning Strategy
  10. Posterior Distributions
  11. TS and Optimization
  12. Offline Evaluation
  13. Online Experiments
  14. Conclusion

Figures (3)

  • Figure 1: Scatter plots of the log of average number of clicks received and the log of budget allocated for various ad line groups with distinct advertisers' industries, channels, supply sources, and audiences.
  • Figure 2: Graphical representation of model (\ref{['eqn:hierachical_model']}). Red nodes are the selected base arm at round $t$.
  • Figure 3: Simulation results on Amazon's campaign data, averaged over 100 random seeds. Shaded areas represent the 95% CI.