What does this research mean for the field?

The combined effect of popularity and positivity bias in recommender systems disproportionately concentrates positivity bias on popular items, which can be effectively mitigated using a percentile-based rating transformation pre-processing strategy without significant accuracy loss. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To evaluate how the combined impact of popularity and positivity biases (multifactorial bias) affects item-side exposure fairness and to develop an efficient data-driven mitigation strategy.

August 22, 2026ACM Transactions on Information SystemsOpen Access

The Unfairness of Multifactorial Bias in Recommendation

View Full Paper

Ask AI

Bookmark

Authors

MMMasoud Mansoury JHJin Huang MPMykola Pechenizkiy

Discussion

Member takes

Overview

Simulation study reveals compounding effects of popularity and positivity biases in recommender systems, suggesting percentile-based transformations enhance exposure fairness efficiently.

Key Points

To evaluate how the combined impact of popularity and positivity biases (multifactorial bias) affects item-side exposure fairness and to develop an efficient data-driven mitigation strategy.
Conducted simulation experiments to assess how positivity bias concentrates across popular versus unpopular items.
Implemented a percentile-based rating transformation pre-processing method.
Evaluated the transformation across six recommendation algorithms on four public datasets, both independently and integrated with post-processing fairness pipelines.
Simulations demonstrated that positivity bias disproportionately concentrates on popular items, significantly exacerbating their over-exposure in recommendations.
Percentile-based rating transformation improved exposure fairness with negligible loss in recommendation accuracy across all six algorithms and four datasets.
Integrating pre-processing into post-processing pipelines yielded comparable or superior fairness while reducing overall computational overhead.

Cite This Study

Mansoury et al. (2026) studied this question.

synapsesocial.com/papers/6a895e6bca7ade938187c8d1 https://doi.org/10.1145/3841477

View Full Paper

Ask AI

Bookmark

August 22, 2026ACM Transactions on Information SystemsOpen Access

The Unfairness of Multifactorial Bias in Recommendation

View Full Paper

Ask AI

Bookmark

Authors

MMMasoud Mansoury JHJin Huang MPMykola Pechenizkiy

Discussion

Member takes

Overview

Simulation study reveals compounding effects of popularity and positivity biases in recommender systems, suggesting percentile-based transformations enhance exposure fairness efficiently.

Key Points

To evaluate how the combined impact of popularity and positivity biases (multifactorial bias) affects item-side exposure fairness and to develop an efficient data-driven mitigation strategy.
Conducted simulation experiments to assess how positivity bias concentrates across popular versus unpopular items.
Implemented a percentile-based rating transformation pre-processing method.
Evaluated the transformation across six recommendation algorithms on four public datasets, both independently and integrated with post-processing fairness pipelines.
Simulations demonstrated that positivity bias disproportionately concentrates on popular items, significantly exacerbating their over-exposure in recommendations.
Percentile-based rating transformation improved exposure fairness with negligible loss in recommendation accuracy across all six algorithms and four datasets.
Integrating pre-processing into post-processing pipelines yielded comparable or superior fairness while reducing overall computational overhead.

Cite This Study

Mansoury et al. (2026) studied this question.

synapsesocial.com/papers/6a895e6bca7ade938187c8d1 https://doi.org/10.1145/3841477

View Full Paper

Ask AI

Bookmark