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 recommendation 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.
Figures & tables
Figure 1: Recommendation AUC plotted against Representation AUC for synthetic datasets with different ϵ parameters. VAE and VAERel .
Figure 2: Gender and Age Recommendation AUC plotted against Representation AUC for the Movielens 1M datasets with different model parameters. VAERel and VAEAfrl* .
Figure 3: Gender and Age Recommendation AUC plotted against Representation AUC for the Movielens 1M datasets with different model parameters. VAE2adv and VAEAfrl* .
Figure 4: Recommendation AUC plotted against Representation AUC in a synthetic dataset with ϵ=0.74 parameters.
Figure 5: Recommendation metrics plotted for different models and dataset ϵ -parameters.
Figure 6: Comparison of VAE and VAERel Demographic Ratio AUC plotted for each dataset ϵ -parameters.
Figure 7: Recommendation metrics plotted for different models and number of users in dataset.
Figure 8: Recommendation metrics plotted for different models and dataset minority ratio.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 9: Item Ratio and Kendall-Tau plotted against Representation AUC. Comparison of VAE and VAERel on synthetic datasets with different ϵ parameters.
Figure 10: Recommendation AUC plotted against Representation AUC in synthetic datasets with three different ϵ parameters.
Fairness audits for LLM-based recommenders have largely focused on observable outputs, implicitly assuming that stable recommendations reflect stable internal processing. We challenge this assumption with FairGap, the first benchmark to jointly evaluate recommendation fairness at two levels: observable output shift (OBS) and hidden representation shift (IBS), measured through controlled counterfactual identity probes across gender, age, and race. Their relationship is summarized via Representation-Output Alignment (ROA), with quadrant diagnostics for identifying user-level hidden-output mismatch. Applied to six open-weight LLM families across three domains, FairGap reveals pervasive hidden-output decoupling: ROA rarely exceeds 0.22, and a non-negligible user population shows stable outputs despite substantial internal shifts, a mode that output-only audits cannot detect by design. Further, activation steering that reduces IBS by up to 8x simultaneously worsens OBS, demonstrating a fundamental tension between internal and output-level fairness that existing frameworks are unequipped to diagnose.
Chan Aristella Lu, Arya Fayyazi, Junhao Zhang +6
University of Georgia · University of Southern California · Carnegie Mellon University +1
While reasoning rerankers, such as Rank1, have demonstrated strong abilities in improving ranking relevance, it is unclear how they perform on other retrieval qualities such as fairness. We conduct the first systematic comparison of fairness between reasoning and non-reasoning rerankers. Using the TREC 2022 Fair Ranking Track dataset, we evaluate six reranking models across multiple retrieval settings and demographic attributes. Our findings demonstrate reasoning neither improve nor harm fairness compared to non-reasoning approaches. Our fairness metric, Attention-Weighted Rank Fairness (AWRF) remained stable (0.33-0.35) across all models, even as relevance varies substantially (nDCG 0.247-1.000). Demographic breakdown analysis revealed fairness gaps for geographic attributes regardless of model architecture. These results indicate that future work in specializing reasoning models to be aware of fairness attributes could lead to improvements, as current implementations preserve the fairness characteristics of their input ranking.
Saron Samuel, Benjamin Van Durme, Eugene Yang
Johns Hopkins University · Human Language Technology Center of Excellence
Fayyazi et al. (2025) recently proposed FACTER, a model-agnostic framework designed to jointly enforce fairness and statistical coverage in LLM-based recommendation through conformal thresholding and iterative prompt repair. In this work, we conduct a reproducibility study of the FACTER framework across diverse architectures and dataset sparsity levels, evaluating both the original open-ended generation task and a constrained re-ranking extension. Under the strict reproduction, we observe a divergence in recommendation utility, which we trace to underspecified target-set evaluation in the original study. We then use the constrained re-ranking setting to evaluate FACTER when the candidate set is fixed, and introduce a static Fair Zero-Shot baseline to isolate the contribution of the iterative prompt repair loop. Our analysis shows that FACTER consistently reduces adaptive-threshold violation counts, but that these reductions are not consistently reflected under the fixed threshold or in global fairness metrics. In the constrained ranking setting, static fairness instructions achieve comparable semantic-parity outcomes to FACTER's dynamic repair loop, suggesting that the additional online repair mechanism provides limited benefit in this formulation. All code and reproduction artifacts are available at https://github.com/oscar-omlf/facter-repr.
Oscar Miró López-Feliu, Daimy van Loo, Xanthos Kekkos +2