# Exploring How Fair Model Representations Relate to Fair Recommendations

arXiv: https://arxiv.org/abs/2603.24396 · HTML: https://mlsanity.com/paper/2603.24396/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 8 of its 15 references that are in the map.
- First posted on arXiv 25 Mar 2026.

## Builds on (most cited first, 8 of 8)

- [Flexibly Fair Representation Learning by Disentanglement](https://mlsanity.com/paper/1906.02589/) (2019)
- [Compositional Fairness Constraints for Graph Embeddings](https://mlsanity.com/paper/1905.10674/) (2019)
- [Beyond Parity: Fairness Objectives for Collaborative Filtering](https://mlsanity.com/paper/1705.08804/) (2017)
- [DeBayes: a Bayesian Method for Debiasing Network Embeddings](https://mlsanity.com/paper/2002.11442/) (2020)
- [Privacy and Fairness in Recommender Systems via Adversarial Training of User Representations](https://mlsanity.com/paper/1807.03521/) (2018)
- [Can We Trust Recommender System Fairness Evaluation? The Role of Fairness and Relevance](https://mlsanity.com/paper/2405.18276/) (2024)
- [Consumer-side Fairness in Recommender Systems: A Systematic Survey of Methods and Evaluation](https://mlsanity.com/paper/2305.09330/) (2023)
- [Adaptive Fair Representation Learning for Personalized Fairness in Recommendations via Information Alignment](https://mlsanity.com/paper/2404.07494/) (2024)

## Abstract (arXiv)

One of the many fairness definitions pursued in recent recommender system research targets mitigating demographic information encoded in model representations. Models optimized for this definition are typically evaluated on how well demographic attributes can be classified given model representations, with the (implicit) assumption that this measure accurately reflects textitrecommendation parity, i.e., how similar recommendations given to different users are. We challenge this assumption by comparing the amount of demographic information encoded in representations with various measures of how the recommendations differ. We propose two new approaches for measuring how well demographic information can be classified given ranked recommendations. Our results from extensive testing of multiple models on one real and multiple synthetically generated datasets indicate that optimizing for fair representations positively affects recommendation parity, but also that evaluation at the representation level is not a good proxy for measuring this effect when comparing models. We also provide extensive insight into how recommendation-level fairness metrics behave for various models by evaluating their performances on numerous generated datasets with different properties.