David Li, Angela Li · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840939
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Tiered service platforms in mobility, delivery, and cloud computing accelerate premium requests, and aggressive prioritization can silently degrade the experience of regular users. We study premium admission as an auditable admission-control problem over streaming telemetry: given per-region arrival rates, service-time moments, and a declared tolerance ? on the mean delay inflation of regular users, how many premium requests may be prioritized? Building on a two-class M/G/1 model with probabilistic gap insertion and head-of-line promotion, we derive a closed-form fairness certificate that maps three telemetry statistics and the platform's prioritization parameters to an admissible premium-to-regular ratio, prove its monotonicity, and show that it is a conservative surrogate of the exact nonpreemptive-priority formula whenever promoted requests wait at most one mean service time. The certificate composes across heterogeneous regions through a per-region cap layer and a shared-slack pooling rule with explicitly declared per-region budgets, and it acts as a state-dependent gate for online controllers. We evaluate FairRank on trace-driven replays of NYC TLC high-volume for-hire vehicle records from three weeks of 2024, comparing nine admission policies, including an established utilization-threshold admission-control baseline, across three premium-candidate definitions and sweeps over tolerance, calibration, scale, and time. Certificate-gated admission reduces mean regular-user delay inflation about five-fold relative to unconstrained prioritization at the evaluated operating points, and the telemetry-updated adaptive variant attains the lowest inflation of all tested policies. Because the certificate is model-conditional, we also isolate which assumptions matter: a synthetic study shows that heavy-tailed service times leave the certificate intact while bursty arrivals roughly double realized inflation, and a diagnostic on the real trace explains the remaining gap and gives operators a conservative parameterization that meets the nominal tolerance in every zone.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.