Efficient Online Marketing Optimization Based on Hierarchical Unimodal Bandit Algorithm

The application of online learning approaches is now widely employed in numerous e-commerce settings, including A/B testing for customer orientation and coupon distribution strategies. A key challenge lies in obtaining an optimal solution swiftly, which corresponds to achieving rapid convergence in cumulative regret. Nevertheless, conventional multi-armed bandit (MAB) algorithms face computational limitations due to regret growth with the number of arms, becoming infeasible as the arm count increases. In this study, we present a modified version of the MAB problem, where arms are grouped into clusters, applicable in situations like joint price and share selection. To tackle this, we design the Two-Level Policy (TLP) algorithm for optimizing joint price and share selection. Moreover, when the reward distribution within each cluster is Unimodal, we introduce the Unimodal Two-Level Policy (UTLP) algorithm. Our numerical tests demonstrate that the proposed algorithms outperform current benchmarks.