Fighting for Fairness: A Re-ranking Approach for Recommendation Systems

Recommendation systems shape user experiences across domains such as e-commerce, job advertising, and entertainment, making fair and balanced recommendations essential. Existing fair re-ranking methods often overlook item-category bias and focus on binary sensitive attributes. We propose a fairness-aware re-ranking approach that uses historical demographic differences as a corrective signal to reduce category-level disparities. Our method supports multiple sensitive attributes, including gender, age, and occupation, and can be applied post-hoc to existing recommenders. Experiments on three real-world datasets show substantial reductions in social bias with little to no loss in recommendation quality.