Post-processing re-ranking is a practical strategy for mitigating demographic bias in recommendation systems. Existing approaches typically enforce group-level category distributions using static, point-estimated preferences at a single granularity, overlooking within-group heterogeneity, ignoring estimation uncertainty for data-sparse groups, and offering limited theoretical guarantees. We propose Adaptive Hierarchical Fairness (AHF), a re-ranking framework that models preference structure and fairness constraints jointly across multiple levels of a category taxonomy. AHF employs a Bayesian hierarchical model that decomposes category affinities into global, group, and individual components, yielding uncertainty-aware corrections that prevent over-adjustment for small groups. We introduce a hierarchical Kullback–Leibler (KL) divergence objective that enforces consumer-side demographic parity across coarse and fine-grained category levels simultaneously, and we prove that controlling per-user KL divergence implies a bounded group-level demographic parity gap, with the bound decomposed into an optimization term and an explicitly controlled target-gap term. For scalable deployment, we devise a locality-sensitive hashing (LSH) sketch that amortizes greedy re-ranking via bucket-level ranking templates with lightweight per-user refinement. Experiments on ML-100K, ML-1M, and Yelp show that AHF reduces category disparity by 31–48% relative to the strongest category-aware baselines while improving NDCG@20 by up to 5.7%. The LSH sketch achieves 22–41 × speedups with under 5% relative fairness degradation. Statistical significance is assessed with Wilcoxon signed-rank tests across users.

Elevators play a pivotal role in modern urban living, boosting productivity and convenience efficiently. In elevator systems, the optimization of Multi-Agent Systems (MAS) is indispensable as it enhances agent coordination, adaptability, delay reduction, client satisfaction, and resource use. In this paper, we introduce an algorithm based on SARL MAS designed to enhance elevator controller performance. Our approach compares Centralized and Distributed Agent Systems, demonstrating the superiority of Distributed Agent Systems due to their improved responsiveness, efficiency, and adaptability. Our findings provide valuable insight into the use of SARL MAS not only for elevator control but also for other applications such as queue management systems and resource allocation in computing, highlighting the benefits of a distributed approach.