FEIR: Quantifying and Reducing Envy and Inferiority for Fair Recommendation of Limited Resources
Автор: ACM RecSys
Загружено: 2025-10-01
Просмотров: 14
Описание: The speaker addresses fairness in recommendation when items have limited availability, where users must compete. They critique top-k recommendation for maximizing utility while ignoring competition, which wastes users’ effort on unattainable items. The work introduces inferiority as an individual-level metric of competitive disadvantage and combines it with the existing envy metric. FEIR converts utility, envy, and inferiority into differentiable probabilistic objectives and optimizes a multi-objective loss via gradient descent. The method is model-agnostic post-processing, taking any recommender’s scores and adjusting assignment probabilities. Experiments on synthetic and real data plot Pareto frontiers, where FEIR consistently dominates baselines. The approach encourages competition-aware allocations of jobs and education resources.
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
