Adaptive Privacy-Aware Fair Exposure Allocation for AI-Generated Content in Decentralized Advertising Ecosystems

Abstract

Decentralized advertising ecosystems distribute content creation, delivery, and measurement across many stakeholders. The rapid growth of AI-generated content (AIGC) intensifies compe- tition for limited attention while privacy regulation and user expectations restrict fine-grained tracking. This paper develops a unified framework for adaptive exposure allocation that bal- ances three objectives: platform efficiency (conversion or revenue), fairness of exposure across creators and content cohorts, and explicit privacy protection. We formalize allocation as an online optimization problem with fairness constraints and differential privacy budgets. We pro- pose an adaptive algorithm that combines bandit-style exploration, privacy-preserving aggregate feedback, and projection onto fairness-feasible sets. We provide regret and fairness-violation guarantees under privacy noise, and demonstrate empirical gains on synthetic and semi-realistic ad logs: compared to strong baselines, our method achieves higher utility at the same privacy level and significantly reduces exposure concentration among head content. The framework is modular and compatible with decentralized reporting and incentive mechanisms.