Paper Information | SBC BOOKS

Language (*)

Title (*)

Keywords (*)

Abstract (*)

Recommender systems are increasingly required to satisfy fairness and transparency constraints. However, the interaction between calibration-based fairness interventions and recommendation explainability remains insufficiently understood. This paper proposes a calibration-driven framework that extends single-category calibration to a multi-category setting, enabling a richer representation of user preferences across item dimensions. The approach integrates fairness-aware calibration with post-hoc knowledge-graph-based explanations into a unified pipeline. We evaluate both recommendation performance and explanation quality, considering alignment with user profiles, accuracy, miscalibration, and popularity bias, as well as explanation metrics such as recency, popularity, and diversity. Results show improved alignment, reduced miscalibration and popularity bias, and preserved accuracy, while explanation quality remains stable. These findings indicate that calibration-based fairness can be achieved without compromising explainability.

Pages (*)

File Link

Title

Keywords

Abstract

Recommender systems are increasingly required to satisfy fairness and transparency constraints. However, the interaction between calibration-based fairness interventions and recommendation explainability remains insufficiently understood. This paper proposes a calibration-driven framework that extends single-category calibration to a multi-category setting, enabling a richer representation of user preferences across item dimensions. The approach integrates fairness-aware calibration with post-hoc knowledge-graph-based explanations into a unified pipeline. We evaluate both recommendation performance and explanation quality, considering alignment with user profiles, accuracy, miscalibration, and popularity bias, as well as explanation metrics such as recency, popularity, and diversity. Results show improved alignment, reduced miscalibration and popularity bias, and preserved accuracy, while explanation quality remains stable. These findings indicate that calibration-based fairness can be achieved without compromising explainability.

#	Name
1	Paul Atauchi(paul.atauchi@usp.br)
2	Andre Zanon(andrelzanon10@gmail.com)
3	Leonardo Rocha(lcrocha@ufsj.edu.pr)
4	Marcelo Manzato(manzato@icmc.usp.br)

Paper Registration

Fill in your paper information

English Information

Authors

Reference

#	Reference
No reference registered