Candidate Retrieval for Provider Fairness under Severe Recommendation Imbalance

Abstract

Provider fairness in recommender systems is commonly addressed through reranking, which assumes that the candidate set already contains sufficient items from under-exposed provider groups. We study a setting in which this assumption breaks down: catalog representation does not translate into recommendation exposure, and minority items are largely missing from the candidates passed to the reranker. In this regime, methods such as FA*IR, calibration-based reranking, or minority boosting may fail, as suitable items are simply absent on their inputs. We address this bottleneck with FairSAE, an activation-steering method that, prior to candidate retrieval, shifts user representations toward minority items through a contrastive direction in a latent space induced by a sparse autoencoder (SAE). This intervention surfaces minority items from the full catalog and can be combined with existing fairness-aware rerankers. Across the MovieLens-25M and Steam datasets, FairSAE-steered retrieval substantially improves the fairness-accuracy trade-off compared to standalone reranking. These results show that provider fairness under severe recommendation imbalance depends not only on reranking, but also on which items are retrieved in the first place. See https://osf.io/nydez for codes and data.