A Reproducibility Study of Product-side Fairness in Bundle Recommendation
Автор: ACM RecSys
Загружено: 2025-10-02
Просмотров: 71
The speaker examines fairness in recommender systems for bundles, focusing on exposure allocated to products. A multi-level evaluation framework measures exposure disparity at bundle and item levels using recall, NDCG, and fairness metrics such as Gini/SSG. Experiments on music, books, and fashion datasets with several state-of-the-art models reveal amplified popularity bias and differing frequency distributions across domains. Results show no single model excels across accuracy and fairness simultaneously. Bundle-level fairness can mask item-level unfairness, necessitating multi-faceted evaluation. User behavior groups (bundle- vs item-preferring) affect fairness outcomes. Future work targets optimizing multi-view aggregation and expanding fairness dimensions.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
