Optimizing Advertising Efficacy: Implementing Cost-Effective Multi-Armed Bandit Algorithms

This paper proposes a novel approach of multi-armed bandit (MAB) algorithms for online advertising across multiple platforms with no budget limitation. The study adapts the application of two algorithms, Cost-Subsidized Upper Confidence Bound Bandit (CS-UCB) and Cost-aware Cascading Upper Confidence Bound Bandit (CC-UCB), with the objective to maximize the click-through rate (CTR) while considering cost efficiency. Departing from traditional budget-constrained models such as Bandits with Knapsacks (BwK), this study assumes a competitive advertising environment where market share and ad exposure are prioritized over strict budget adherence and cost control. It provides advertisers flexibility in cost control, maximizing CTR within the bounds of acceptable cost concessions, balancing the dual objectives of minimize quality regret and cost regret. The introduction of this novel appraoch extends the application of Multi-Armed Bandit (MAB) algorithms to advertising strategies in highly competitive markets without budget constraints. This study analysises the performance of both CC-UCB and CS-UCB algorithms with empricial research by using real-world data, showing online advertisements bidding strategies from multiple ad platforms in real-time bidding (RTB) system.