EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Controllable Many-Objective Optimization for Fairness-Aware Uplift Modeling in Recommender Systems

Sunday O. Oladejo

IEEE Access · 2026 · 14 · 109905-109933

DOI: 10.1109/access.2026.3709619 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Sunday O. Oladejo. Die Arbeit nennt 75 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Division for Research Development, Stellenbosch University, South Africa;, Standard Bank Lab for Recommender Systems.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper investigates the distribution of promotional offers under simultaneous budget, fairness, and treatment overlap constraints, using causal uplift modeling to estimate the incremental impact on personalized platforms and, specifically, recommender systems. The promotional assignment is formulated as first a Multi-Choice Knapsack Problem (MCKP), and then a Many-Objective Optimization framework, which constitutes a key novelty over conventional single-objective or fixed-weight methods by enabling flexible management of competing goals, such as fairness, budget adherence, overlap control, and uplift maximization across groups to prevent systemic under-allocation to under-served groups while maximizing the revenue impact. Machine learning models–that is, S-/ T-/ X-Learners and Causal Forests–are employed as a data-driven mechanism for predicting the treatment impact estimates. Owing to the non-convexity and conflicting characteristics of the objective functions, this paper adopts the Nondominated Sorting Genetic Algorithm III (NSGA-III) and Multiobjective Evolutionary Algorithm Based on Decomposition (MOEA/D), both advanced algorithms designed to solve complex multi-objective and many-objective optimization problems. The proposed frameworks are evaluated using the Hillstrom datasets via extensive Monte Carlo simulations and the use of statistical tests such as the Wilcoxon and Friedman tests. The respective proposed solutions are discussed alongside their trade-offs. The code and data for this study are openly available at: https://github.com/DayoSun/Controllable-Many-Objective-Optimization-for-Fairness-Aware-Uplift-Modeling-in-Recommender-Systems

Förderung

  • Division for Research Development
  • Stellenbosch University, South Africa;
  • Standard Bank Lab for Recommender Systems

Zitieren

APA

Oladejo, S. O. (2026). Controllable Many-Objective Optimization for Fairness-Aware Uplift Modeling in Recommender Systems. IEEE Access, 14, 109905-109933. https://doi.org/10.1109/access.2026.3709619

Vancouver

Oladejo SO. Controllable Many-Objective Optimization for Fairness-Aware Uplift Modeling in Recommender Systems. IEEE Access. 2026;14:109905-109933. doi:10.1109/access.2026.3709619

BibTeX
@article{Oladejo2026,
  author = {Oladejo, Sunday O.},
  title = {Controllable Many-Objective Optimization for Fairness-Aware Uplift Modeling in Recommender Systems},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {109905--109933},
  doi = {10.1109/access.2026.3709619}
}
Kostenlos starten →